Search This Blog

Tuesday, December 9, 2025

Climate change in Africa

From Wikipedia, the free encyclopedia
Graph showing temperature change in Africa between 1901 and 2021, with red colour being warmer and blue being colder than average (The average temperature during 1971–2000 is taken as the reference point for these changes.)

Climate change in Africa is a serious threat as Africa is one of the most vulnerable regions to the effects of climate change, despite contributing the least to causing it. Climate change is causing increasingly erratic rainfall patterns, more frequent extreme weather events including droughts, floods, and rising sea surface temperatures in Africa. These changes threaten food and water security, biodiversity, public health, and economic development. Africa is currently warming faster than the rest of the world on average.

Climate change intensifies existing socioeconomic vulnerabilities. Large segments of the African population depend on climate-sensitive livelihoods such as agriculture (55 - 62% of the workforce in sub-Saharan Africa)[4] and already live in poverty, heightening their exposure to shocks. Health outcomes worsen as heat stress, vector-borne diseases (such as malaria and dengue), and malnutrition become more prevalent. Over half (56%) of the over 2,000 recorded public health incidents in Africa between 2001 and 2021 were connected to climate change. Resource scarcity contributes to displacement and conflict, particularly in fragile regions. Urban areas, often characterized by informal settlements, face heightened risks from flooding and extreme heat.

Agriculture is one of the most vulnerable sectors, as most African farmers rely on rainfed crops. Reduced and unpredictable rainfall, combined with higher temperatures, drives soil moisture loss, desertification (especially in the Sahara) and shifts suitable growing areas. These changes lower yields of staple crops, undermining food security and worsening hunger. Livestock health is increasingly compromised by heat stress and shifting disease patterns. Coastal and marine ecosystems face warming seas and rising levels, which threaten fisheries and densely populated coastal settlements.

The economic toll of climate change is severe. On average African countries face climate-related losses amounting to 2-5% of GDP annually, while adaptation costs in sub-Sahran Africa are projected at USD 30-50 billion per year over the next decade. This threatens development gains and places pressure on governments and international institutions to mobilise climate finance.

Africa's climate change adaptation strategies focus on building resilience through climate-smart agriculture, sustainable water management, ecosystem conservation, and strengthening health and infrastructure systems. These approaches prioritise enhancing governance, mobilising climate finance and investment, and fostering community participation to address vulnerability holistically. Continental and national frameworks emphasise multi-sectoral coordination, technology adoption, and capacity building to support sustainable development and reduce climate risks.

Greenhouse gas emissions

Africa's per person greenhouse gas emissions are low compared to other continents. Emissions from land use change are uncertain, especially in Central Africa. The main source of uncertainty comes from carbon dioxide fluxes in the LULUCF sector (this acronym stands for land use, land-use change, and forestry).

Impacts

Temperature and weather changes

Köppen climate classification map for Africa for 1991–2020
 
2071–2100 map under the most intense climate change scenario. Mid-range scenarios are currently considered more likely.

Observed surface temperatures have generally increased over Africa since the late 19th century to the early 21st century by about 1 °C, but locally as much as 3 °C for minimum temperature in the Sahel at the end of the dry season. Observed precipitation trends indicate spatial and temporal discrepancies as expected. The observed changes in temperature and precipitation vary regionally.

Current climate models (as summarised in the IPCC Sixth Assessment Report) predict increases in frequency and intensity of drought and heavy rainfall events. They also predict decreases in mean precipitation almost everywhere in Africa, with medium to high confidence. However, local rainfall trends and socio-climatic interactions are likely to manifest in mixed patterns. Therefore, the converging impacts of climate change will vary across the continent. In rural areas, rainfall patterns influence water usage.

A study in 2019 predicted increased dry spell length during wet seasons and increased extreme rainfall rates in Africa. In other words: "both ends of Africa's weather extremes will get more severe". The research found that most climate models will not be able to capture the extent of these changes because they are not convection-permitting at their coarse grid scales.

Sea level rise

Aerial view of the Tanzanian capital Dar es Salaam

In Africa, future population growth amplifies risks from sea level rise. Some 54.2 million people lived in the highly exposed low elevation coastal zones (LECZ) around 2000. This number will effectively double to around 110 million people by 2030. By 2060 it will be around 185 to 230 million people, depending on the extent of population growth. The average regional sea level rise will be around 21 cm by 2060. At that point climate change scenarios will make little difference. But local geography and population trends interact to increase the exposure to hazards like 100-year floods in a complex way.

Abidjan, the economic powerhouse of Ivory Coast
Populations within 100-year floodplains.
Country 2000 2030 2060 Growth 2000–2060
Egypt 7.4 13.8 20.7 0.28
Nigeria 0.1 0.3 0.9 0.84
Senegal 0.4 1.1 2.7 0.76
Benin 0.1 0.6 1.6 1.12
Tanzania 0.2 0.9 4.3 2.3
Somalia 0.2 0.6 2.7 1.7
Côte d'Ivoire 0.1 0.3 0.7 0.65
Mozambique 0.7 1.4 2.5 0.36
A man looking out over the beach from a building destroyed by high tides in Chorkor, a suburb of Accra. Sunny day flooding caused by sea level rise, increases coastal erosion that destroys housing, infrastructure and natural ecosystems. A number of communities in Coastal Ghana are already experiencing the changing tides.

In the near term, some of the largest displacement is projected to occur in the East Africa region. At least 750,000 people there are likely to be displaced from the coasts between 2020 and 2050. Scientific studies estimate that 12 major African cities would collectively sustain cumulative damages of US$65 billion for the "moderate" climate change scenario RCP4.5 by 2050. These cities are Abidjan, Alexandria, Algiers, Cape Town, Casablanca, Dakar, Dar es Salaam, Durban, Lagos, Lomé, Luanda and Maputo. Under the high-emission scenario RCP8.5 the damage would amount to US$86.5 billion. The version of the high-emission scenario with additional impacts from high ice sheet instability would involve up to US$137.5 billion in damages. The damage from these three scenarios accounting additionally for "low-probability, high-damage events" would rise to US$187 billion, US$206 billion and US$397 billion respectively. In these estimates, the Egyptian city of Alexandria alone accounts for around half of this figure. Hundreds of thousands of people in its low-lying areas may already need relocation in the coming decade. Across sub-Saharan Africa as a whole, damage from sea level rise could reach 2–4% of GDP by 2050. However this figure depends on the extent of future economic growth and adaptation.

The remains of Leptis Magna amphitheater, with the sea visible in the background

In the longer term, Egypt, Mozambique and Tanzania are likely to have the largest number of people affected by annual flooding amongst all African countries. This projection assumes global warming will reach 4 °C by the end of the century. That rise is associated with the RCP8.5 scenario. Under RCP8.5, 10 important cultural sites would be at risk of flooding and erosion by the end of the century. These are the Casbah of Algiers, Carthage Archaeological site, Kerkouane, Leptis Magna Archaeological site, Medina of Sousse, Medina of Tunis, Sabratha Archaeological site, Robben Island, Island of Saint-Louis and Tipasa. A total of 15 Ramsar sites and other natural heritage sites would face similar risks. These are Bao Bolong Wetland Reserve, Delta du Saloum National Park, Diawling National Park, Golfe de Boughrara, Kalissaye, Lagune de Ghar el Melh et Delta de la Mejerda, Marromeu Game Reserve, Parc Naturel des Mangroves du Fleuve Cacheu, Seal Ledges Provincial Nature Reserve, Sebkhet Halk Elmanzel et Oued Essed, Sebkhet Soliman, Réserve Naturelle d'Intérêt Communautaire de la Somone, Songor Biosphere Reserve, Tanbi Wetland Complex and Watamu Marine National Park.

Socioeconomic impacts

Survey results from 2022 show that access to wood and water appears to be severely impacted by climate change in Kenya and Cameroon.

Climate change will increasingly impact Africa due to many factors. These impacts are already being felt and will increase in magnitude if action is not taken to reduce global carbon emissions. The impacts include higher temperatures, drought, changing rainfall patterns, and increased climate variability. These conditions have a bearing on energy production and consumption. The recent drought in many African countries, which has been linked to climate change, adversely affected both energy security and economic growth across the continent.

Africa will be one of the regions most impacted by the adverse effects of climate change. Reasons for Africa's vulnerability are diverse and include low levels of adaptive capacity, poor diffusion of technologies and information relevant to supporting adaptation, and high dependence on agro-ecosystems for livelihoods. Many countries across Africa are classified as Least-Developed Countries (LDCs) with poor socio-economic conditions, and by implication are faced with particular challenges in responding to the impacts of climate change.

Pronounced risks identified for Africa in the IPCC's Fifth Assessment Report relate to ecosystems, water availability, and agricultural systems, with implications for food security.

In 2022, over 6,000 respondents from ten African nations took part in a climate survey conducted by the European Investment Bank. The survey found that 88% of respondents claimed climate change was hurting their lives, while 61% of respondents claimed that environmental destruction has impacted their income or source of livelihood. These losses are usually the result of severe drought, increasing sea levels or coastal erosion, or extreme weather events like floods or storms.

More than half of African respondents (57%) said that they or people they know have already made steps to adapt to the effects of climate change. Among these measures are investments in water-saving devices to mitigate the effects of drought and drain clearance ahead of flooding. 34% of all African respondents said climate change is one of the most pressing issues confronting their country, among other key issues such as inflation and access to health care.

Economic impacts

Africa is warming faster than the rest of the world on average. Large portions of the continent may become uninhabitable as a result and Africa's gross domestic product (GDP) may decline by 2% as a result of a 1 °C rise in average world temperature, and by 12% as a result of a 4 °C rise in temperature. Crop yields are anticipated to drastically decrease as a result of rising temperatures and it is anticipated that heavy rains would fall more frequently and intensely throughout Africa, increasing the risk of floods.

Additionally, Africa loses between $7 billion and $15 billion a year due to climate change, projected to reach up to $50 billion by 2030.

Agriculture

Agriculture is a particularly important sector in Africa, contributing towards livelihoods and economies across the continent. On average, agriculture in Sub-Saharan Africa contributes 15% of the total GDP. Africa's geography makes it particularly vulnerable to climate change, and 70% of the population rely on rain-fed agriculture for their livelihoods. Smallholder farms account for 80% of cultivated lands in Sub-Saharan Africa. The IPCC in 2007 projected that climate variability and change would severely compromise agricultural productivity and access to food. This projection was assigned "high confidence". Cropping systems, livestock and fisheries will be at greater risk of pest and diseases as a result of future climate change. Crop pests already account for approximately 1/6th of farm productivity losses. Climate change will accelerate the prevalence of pests and diseases and increase the occurrence of highly impactful events. The impacts of climate change on agricultural production in Africa will have serious implications for food security and livelihoods. Between 2014 and 2018, Africa had the highest levels of food insecurity in the world.

In relation to agricultural systems, heavy reliance on rain-fed subsistence farming and low adoption of climate smart agricultural practices contribute to the sector's high levels of vulnerability. The situation is compounded by poor reliability of, and access to, climate data and information to support adaptation actions. Observed and projected disruptions in precipitation patterns due to climate change are likely to shorten growing seasons and affect crop yield in many parts of Africa. Furthermore, the agriculture sector in Africa is dominated by smallholder farmers with limited access to technology and the resources to adapt.

Climate variability and change have been and continue to be the principal source of fluctuations in global food production across developing countries where production is highly rain-dependent. The agriculture sector is sensitive to climate variability, especially the inter-annual variability of precipitation, temperature patterns, and extreme weather events (droughts and floods). These climatic events are predicted to increase in the future and are expected to have significant consequences to the agriculture sector. This would have a negative influence on food prices, food security, and land-use decisions. Yields from rainfed agriculture in some African countries could be reduced by up to 50% by 2020. To prevent the future destructive impact of climate variability on food production, it is crucial to adjust or suggest possible policies to cope with increased climate variability. African countries need to build a national legal framework to manage food resources in accordance with the anticipated climate variability. However, before devising a policy to cope with the impacts of climate variability, especially to the agriculture sector, it is critical to have a clear understanding of how climate variability affects different food crops. This is particularly relevant in 2020 due to the severe invasion of Locusts adversely affecting agriculture in eastern Africa. The invasion was partially attributed to climate change – the warmer temperature and heavier rainfall which caused an abnormal increase in the number of locusts.

In East Africa, climate change is anticipated to intensify the frequency and intensity of drought and flooding, which can have an adverse impact on the agricultural sector. Climate change will have varying effects on agricultural production in East Africa. Research from the International Food Policy Research Institute (IFPRI) suggest an increase in maize yields for most East Africa, but yield losses in parts of Ethiopia, Democratic Republic of Congo (DRC), Tanzania and northern Uganda. Projections of climate change are also anticipated to reduce the potential of the cultivated land to produce crops of high quantity and quality.

Climate change in Kenya is expected to have large impacts on the agricultural sector, which is predominantly rain-fed and thus highly vulnerable to changes in temperature and rainfall patterns, and extreme weather events. Impacts are likely to be particularly pronounced in the arid and semi-arid lands (ASALs) where livestock production is the key economic and livelihood activity. In the ASALs, over 70% of livestock mortality is a result of drought. Over the next 10 years, 52% of the ASAL cattle population are at risk of loss because of extreme temperature stress.

Climate change will exacerbate the vulnerability of the agricultural sector in most Southern African countries which are already limited by poor infrastructure and a lag in technological inputs and innovation. Maize accounts for nearly half of the cultivated land in Southern Africa, and under future climate change, yields could decrease by 30%. Temperatures increases also encourage a wide spread of weeds and pests.

Climate change will significantly affect agriculture in West Africa by increasing the variability in food production, access and availability.

Higher rainfall intensity, prolonged dry spells and high temperatures are expected to negatively impact cassava, maize and bean production in Central Africa. Floods and erosion occurrence are expected to damage the already limited transportation infrastructure in the region leading to post harvest losses. Exportation of economic crops like coffee and cocoa are on the rise within the region but these crops are highly vulnerable to climate change. Conflicts and political instability have had an impact on agriculture contribution to the regional GDP and this impact will be exacerbated by climatic risks.

Africa's gross domestic product (GDP) may decline by 2% as a result of a 1 °C rise in average world temperature, and by 12% as a result of a 4 °C rise in temperature. Crop yields are anticipated to drastically decrease as a result of rising temperatures and an increase in the likelihood of drought throughout the continent. Additionally, it is anticipated that heavy rains would fall more frequently and intensely throughout Africa, increasing the risk of floods.

Energy

Solar lighting and electricity installed in the home of a Tanzanian woman.

With increasing population and corresponding energy demand, energy security must be addressed because energy is crucial for sustainable development. Climate change has affected energy sectors in Africa as many countries depend on hydropower generation. Decreasing rainfall levels and droughts have resulted in lower water levels in dams with adverse impacts on hydropower generation. This has resulted in low electrical energy production, high cost of electricity and power outages or load-shedding in some African countries that depend on hydroelectric power generation. Disruptions in hydropower generation have negatively affected various sectors in countries such as Ghana, Uganda, Kenya, and Tanzania.

Water scarcity

Water shortage in Ethiopia.

Water quality and availability have deteriorated in most areas of Africa, particularly due to climate change. Water resources are vulnerable and have the possibility of being strongly impacted by climate change with vast ramifications on human societies. The IPCC predicts millions of people in Africa will persistently face increased water stress due to climate variability and change (IPCC 2013). Changes in precipitation patterns directly affect surface runoff and water availability.

Climate change is likely to further exacerbate water-stressed catchments across Africa – for example the Rufiji basin in Tanzania – owing to diversity of land uses, and complex sociopolitical challenges.

Health impacts

African countries have the least efficient public health systems in the world. Infectious disease burdens such as malaria, schistosomiasis, dengue fever, meningitis, which are sensitive to climate impacts, are highest in the sub-Saharan African region. For instance, over 90 percent of annual global malaria cases are in Africa. Changes in climate will affect the spread of infectious agents as well as alter people's disposition to these infections.

According to the IPCC's Sixth Assessment Report, climate change poses a significant threat to the health of tens of millions of Africans, as it exposes them to non-optimal temperatures, extreme weather, and an increased range and transmission rate of infectious diseases.

Climate change, and resulting in increased temperatures, storms, droughts, and rising sea levels, will affect the incidence and distribution of infectious disease across the globe.

In July 2021, the World Food Programme (WFP) blamed the ongoing southern Madagascar food crisis as being caused solely by climate change and not by war or conflict. It was declared to be first famine caused by climate change.

Malaria

In Africa malaria continues to have dramatic effects on the population. As climate change continues, the specific areas likely to experience the year-round, high-risk transmission of malaria will shift from coastal West Africa to an area between the Democratic Republic of the Congo and Uganda, known as the African Highlands.

Scientific limitations when examining shifting malaria transmission rates in the African Highlands are similar to those related to broader understandings of climate change and malaria. While modeling with temperature changes shows that there is a relationship between an increase in temperature and an increase in malaria transmission, limitations still exist. Future population shifts that affect population density, as well as changes in the behavior of mosquitos, can affect transmission rates and are limiting factors in determining the future risk of malaria outbreaks, which also affect planning for correct outbreak response preparation.

With regards to malaria transmission rates in the African Highlands, factors and exposures resulting from drastic environmental changes like warmer climates, shifts in weather patterns, and increases in human impact such as deforestation, provide appropriate conditions for malaria transmission between carrier and host. Specifically, malaria is caused by the Plasmodium falciparum and Plasmodium vivax parasites which are carried by the vector Anopheles mosquito. Even though the Plasmodium vivax parasite can survive in lower temperatures, the Plasmodium falciparum parasite will only survive and replicate in the mosquito when climate temperatures are above 20 °C. Increases in humidity and rain also contribute to the replication and survival of this infectious agent. Exposure to malaria will become a greater risk to humans as the number of female Anopheles mosquitos infected with either the Plasmodium falciparum or Plasmodium vivax parasite increases.

Studies show an overall increase in climate suitability for malaria transmission resulting in an increase in the population at risk of contracting the disease. Of significant importance is the increase of epidemic potential at higher altitudes (like the African Highlands). Rising temperatures in these areas have the potential to change normally non-malarial areas to areas with seasonal epidemics. Consequently, new populations will be exposed to the disease resulting in healthy years lost. In addition, the disease burden may be more detrimental to areas that lack the ability and resources to effectively respond to such challenges and stresses.

As climate change shifts geographic areas of transmission to the African Highlands, the challenge will be to find and control the vector in areas that have not seen it before.

Impacts on conflicts and migration

The United Nations Environment Programme produced a post-conflict environmental assessment of Sudan in 2007. According to this report, environmental stresses in Sudan are interlinked with other social, economic and political issues, such as population displacement and competition over natural resources. Regional climate change, through decreased precipitation, was thought to have been one of the factors which contributed to the conflict in Darfur. Along with other environmental issues, climate change could negatively affect future development in Sudan. One of the recommendations made by UNEP was for the international community to assist Sudan in adapting to climate change.

Impacts by region

Central Africa

Central Africa, for the most part, is landlocked and is geographically threatened by climate change. Due to its high climate variability and rainfed agriculture, Central Africa is expected to experience longer and more frequent heatwaves as well as an increase in wet extremes. The global mean temperature in this region is to increase by 1.5 °C to 2 °C.

The carbon dioxide-absorbing capacity of forests in the Congo Basin have decreased. This decrease has occurred due to increasing heat and drought causing decreased tree growth. This suggests that even unlogged forests are being affected by climate change. A Nature study indicates that by 2030, the African jungle will absorb 14 percent less carbon dioxide than it did from around 2005–2010, and will absorb none at all by 2035.

Eastern Africa

Situated almost entirely in the tropics, rainfall in Eastern Africa is dominated by the seasonal migration of the tropical-rain band. Eastern Africa is characterized by high spatio-temporal rainfall variability as it spans over 30 degrees of latitude (across the equator). It has influences from both the Indian and Atlantic Oceans, and has major geographic features (highlands) as well as inland water bodies such as Lake Victoria. Therefore the rainfall seasonality varies from a single wet season per year in July–August in parts of the northwest (including Ethiopia and South Sudan, which are meteorologically more connected to West Africa, with the West African monsoon bringing the rains) to a single wet season per year in December – February in the south (over Tanzania), with many areas close to the equator having two rainy seasons per year, approximately in March–May (the "Long Rains") and October to December (the "Short Rains"). Fine-scale variability in rainfall seasonality is often linked to orography and lakes. Inter-annual variability can be large and known controls include variations in Sea surface temperatures (SSTs) of different ocean basins, large-scale atmospheric modes of variability such as the Madden–Julian Osciliation (MJO) and tropical cyclones. The Long Rains are the main crop-growing season in the region. Interannual predictability of this season is low compared to the Short Rains, and recent drying contrasts with climate projections of a wetter future (the "East African climate paradox".).

Eastern Africa has witnessed frequent and severe droughts in recent decades, as well as devastating floods. Trends in rainfall since the 1980s show a general decrease in March – May (MAM) seasonal rains with a slight increase during June – September (JJAS) and October – December (OND) rains, although there appears to have been a recent recovery in the MAM rains. In the future, both rainfall and temperature are projected to change over Eastern Africa. Recent studies on climate projections suggest that average temperature might increase by about 2–3 °C by the middle of the century and 2–5 °C at the end of the century. This will depend on emission scenarios as well as on how the real climate responds compared with the range of possible outcomes shown by models. Climate model projections tend to show an increase in rainfall, particularly during OND season, which is also projected to occur later. This delay in the short rain season, has been linked to the deepening of the Saharan Heat Low under climate change. It should be noted, however that some models predict decreasing rainfall, and for some regions and seasons the very largest rainfall increases predicted have been shown to involve implausible mechanisms due to systematic model errors. In addition, changes of aerosols provide a forcing of rainfall change that is not captured in many assessments of climate projections.

The contrast of the drying trend of MAM (long rains) rainfall in equatorial Eastern Africa, with most models predicting a wetting in the future has been labelled the "East African climate change paradox", although there has been some recent recovery in the rainfall. Studies have shown that the drying trend is unlikely to be purely natural, but may be driven by factors such as aerosols rather than greenhouse gases, further research is needed. The drying has been shown to have been caused by a shorter rainy season, and linked to deepening of the Arabian Heat Low.

Consistent with the uncertainty in rainfall projections, changes in rainy seasons onset are uncertain in equatorial Eastern Africa, although many models predict a later and wetter short rains. The Indian Ocean Dipole (IOD) is known to provide a strong control on inter-annual variability in the short rains, and studies show that extreme IODs may increase under climate change.

Globally, climate change is expected to lead to intensification of rainfall, as extreme rainfall increases at a faster rate with warming than total rainfall does. Recent work shows that across Africa global models are expected to under-estimate the rate of change of this rainfall intensification, and changes in rainfall extremes may be much more widespread than those predicted by global models.

Southern parts of Eastern Africa receive most of their rainfall in a single rainy season during the southern hemisphere's winter: over Tanzania seasonal rainfall is projected to increase under future climate change, although there is uncertainty. Further south, over Mozambique, a shorter season due to a later onset is projected under future climate change, again with some uncertainty.

As an example, Kenya has a high vulnerability to the impacts of climate change. The main climate hazards include droughts and floods as rainfall will likely become more intense and less predictable. Climate models predict that temperatures will rise by 0.5 to 2 °C. In the informal urban settlements of Nairobi the urban heat island effect adds to the problem as it creates even warmer ambient temperatures. This is due to home construction materials, lack of ventilation, sparse green space, and poor access to electrical power and other services.

North Africa

Climate classification maps for the Middle East at present (top) and predicted for North Africa for 2071–2100 under the most intense climate change scenario (bottom). Mid-range scenarios are currently considered more likely.

In 2018, the MENA region emitted 3.2 billion tonnes of carbon dioxide and produced 8.7% of global greenhouse gas emissions (GHG) despite making up only 6% of the global population. These emissions are mostly from the energy sector, an integral component of many Middle Eastern and North African economies due to the extensive oil and natural gas reserves that are found within the region. The Middle East region is one of the most vulnerable to climate change. The impacts include increase in drought conditions, aridity, heatwaves and sea level rise.

Sharp global temperature and sea level changes, shifting precipitation patterns and increased frequency of extreme weather events are some of the main impacts of climate change as identified by the Intergovernmental Panel on Climate Change (IPCC). The MENA region is especially vulnerable to such impacts due to its arid and semi-arid environment, facing climatic challenges such as low rainfall, high temperatures and dry soil. The climatic conditions that foster such challenges for MENA are projected by the IPCC to worsen throughout the 21st century. If greenhouse gas emissions are not significantly reduced, part of the MENA region risks becoming uninhabitable before the year 2100.
 

West Africa and the Sahel

The West African region can be divided into four climatic sub-regions namely the Guinea Coast, Soudano-Sahel, Sahel (extending eastward to the Ethiopian border) and the Sahara, each with different climatic conditions. The seasonal cycle of rainfall is mainly driven by the south-north movement of the Inter-Tropical Convergence Zone (ITCZ) which is characterised by the confluence between moist southwesterly monsoon winds and the dry northeasterly Harmattan.

Based on the inter-annual rainfall variability, three main climatic periods have been observed over the Sahel: the wet period from 1950 to the early 1960s followed by a dry period from 1972 to 1990 and then the period from 1991 onwards which has seen a partial rainfall recovery. During the dry period, the Sahel experienced a number of particularly severe drought events, with devastating effects. The recent decades, have also witnessed a moderate increment in annual rainfall since the beginning of 1990s. However, total annual rainfall remains significantly below that observed during the 1950s.

Some have identified the two recent decades as a recovery period. Others refer to this as a period of "hydrological intensification" with much of the annual rainfall increase coming from more severe rain events and sometimes flooding rather than more frequent rainfall, or similarly other works underline the continuity of the drought even though the rainfall has increased. Since 1985, 54 percent of the population has been affected by five or more floods in the 17 Sahel region countries. In 2012, severe drought conditions in the Sahel were reported. Governments in the region responded quickly, launching strategies to address the issue.

The region is projected to experience changes in rainfall regime, with climate models suggesting that decreases in wet season rainfall are more likely in the western Sahel, and increases more likely in the central to east Sahel, although opposite trends cannot yet be ruled out. These trends will affect the frequency and severity of floods, droughts, desertification, sand and dust storms, desert locust plagues and water shortages.

However, irrespective of the changes in seasonal mean rain, the most intense storms are expected to become more intense, amplifying flood frequency. Enhanced carbon emissions and global warming may also lead to an increase in dry spells especially across the Guinea Coast associated with a reduction of the wet spells under both 1.5 °C and 2 °C global warming level.

Fifteen percent of Sahel region population has also experienced a temperature increase of more than 1 °C from 1970 to 2010. The Sahel region, in particular, will experience higher average temperatures over the course of the 21st century and changes in rainfall patterns, according to the Intergovernmental Panel on Climate Change (IPCC).

Southern Africa

Adaptation

To reduce the impacts of climate change on African countries, adaptation measures are required at multiple scales – ranging from local to national and regional levels. The first generation of adaptation projects in Africa can be largely characterized as small-scale in nature, focused on targeted investments in agriculture and diffusion of technologies to support adaptive decision-making. More recently, programming efforts have re-oriented towards larger and more coordinated efforts, tackling issues that spanning multiple sectors. According to a 2023 study, 59% of African banks have a climate change policy in place, with another 22% planning to implement one. 65% of banks presently consider climate risk when evaluating new clients or projects, with another 23% expecting to do so in the future.

Green finance opportunities and products from surveyed banks in the European Investment Bank's Banking in Africa survey

Improved weather forecasting technology in sub-Saharan Africa is important to inform the response to climate change, to aid decision-making associated with adaptation to climate change for example.

During the 21st Conference of the Parties (COP) in 2015, African heads of state launched the Africa Adaptation Initiative (AAI). The AAI's steering committee is composed of the African Ministerial Conference on Environment (AMCEN) Bureau and the chair of the African Group of Negotiators (AGN).

The Africa Adaptation Initiative is also supported by the European Union. The European Union has partnered with the African Union on the promotion of sustainable resources management, environmental resilience, and climate change mitigation

At the regional level, regional policies and actions in support of adaptation across Africa are still in their infancy. The IPCC's Fifth Assessment Report (AR5) highlights examples of various regional climate change action plans, including those developed by the Southern African Development Community (SADC) and Lake Victoria Basin Committee. At the national level, many early adaptation initiatives were coordinated through National Adaptation Programmes of Action (NAPAs) or National Climate Change Response Strategies (NCCRS). Implementation has been slow however, with mixed success in delivery. Integration of climate change with wider economic and development planning remains limited but growing.

At the subnational level, many provincial and municipal authorities are also developing their own strategies, for example the Western Cape Climate Change Response Strategy. Yet, levels of technical capacity and resources available to implement plans are generally low. There has been considerable attention across Africa given to implementing community-based adaptation projects. There is broad agreement that support to local-level adaptation is best achieved by starting with existing local adaptive capacity, and engaging with indigenous knowledge and practices.

Results regarding African banks' climate risk approach (% of surveyed banks) from the European Investment Bank's Banking in Africa survey 2021

The IPCC highlights a number of successful approaches to promote effective adaptation in Africa, outlining five common principles. These include:

  1. Enhancing support for autonomous forms of adaptation;
  2. Increasing attention to the cultural, ethical, and rights considerations of adaptation (especially through active participation of women, youth, and poor and vulnerable people in adaptation activities);
  3. Combining "soft path" options and flexible and iterative learning approaches with technological and infrastructural approaches (including integration of scientific, local, and indigenous knowledge in developing adaptation strategies)
  4. Focusing on enhancing resilience and implementing low-regrets adaptation options; and
  5. Building adaptive management and encouraging process of social and institutional learning into adaptation activities.

The World Health Organization's report "Adaptation to Climate Change in Africa Plan of Action for the Health Sector 2012–2016" is intended to "provide a comprehensive and evidence-based coordinated response of the health sector to climate change adaptation needs of African countries in order to support the commitments and priorities of African governments." The action plan includes goals like scaling up public health activities, coordinating efforts on an international scale, strengthening partnerships and collaborative efforts, and promoting research on both the effects of climate change as well as effective measures taken in local communities to mitigate climate change consequences.

According to the International Monetary Fund (IMF), Sub-Saharan Africa requires $30–$50 billion in additional financing each year to adapt to the effects of climate change.

Climate financing in the Middle East and North Africa totaled $32.6 billion (2% of the world total) in 2019/2020, while climate investment in Sub-Saharan Africa was $43.8 billion (3% of the global total).

According to the European Investment Bank's Banking in Africa study 2021, African institutions are becoming more conscious of the need to address the dangers posed by climate change and are beginning to capitalize on possibilities in green financing. For example, 54% of questioned banks in the study saw climate change as a strategic concern, and more than 40% had people focusing on climate-related fronts. Sub-Saharan African banks are growing their digital offerings, which has been expedited by the COVID-19 pandemic. The majority of the banks surveyed said that the pandemic has accelerated the speed of digital transformation, and that this shift will be permanent.

The poor and vulnerable are most susceptible, with migrant workers, refugees, and other marginalised groups likely to suffer the most. GDP per capita is not likely to rebound to 2019 levels until 2024, with risks tilting to the downside, and the crisis has reversed a predicted drop in the number of poor people, according to the IMF.

In comparison to pre-crisis forecasts, this might result in an additional 30 million people in Sub-Saharan Africa living in extreme poverty by 2021, as well as an additional nine million in the Middle East and North Africa (MENA) area.

As of 2023, about a third of all African climate funding flows to five major markets: Morocco (7% of African climate investment in 2019/2022), Nigeria (7%), Kenya (7%), Ethiopia (6%), and South Africa (5%).

Over the last decade, worldwide greenfield foreign direct investment has declined at a 3% annual rate, with Africa's global contribution dropping from 12% in 2017 to less than 6% in 2021.

Northern Africa adaptation measures

Climate change specific personnel in surveyed African banks in the European Investment Bank Banking in Africa survey (% of surveyed African banks)

Key adaptations in northern Africa relate to increased risk of water scarcity (resulting from a combination of climate change affecting water availability and increasing demand). Reduced water availability, in turn, interacts with increasing temperatures to create need for adaptation among rainfed wheat production and changing disease risk (for example from leishmaniasis). Most government actions for adaptation centre on water supply side, for example through desalination, inter-basin transfers and dam construction. Migration has also been observed to act as an adaptation for individuals and households in northern Africa. Like many regions, however, examples of adaptation action (as opposed to intentions to act, or vulnerability assessments) from north Africa are limited – a systematic review published in 2011 showed that only 1 out of 87 examples of reported adaptations came from North Africa.

Western Africa adaptation measures

Water availability is a particular risk in Western Africa, with extreme events such as drought leading to humanitarian crises associated with periodic famines, food insecurity, population displacement, migration and conflict and insecurity. Adaptation strategies can be environmental, cultural/agronomic and economic.

Adaptation strategies are evident in the agriculture sector, some of which are developed or promoted by formal research or experimental stations. Indigenous agricultural adaptations observed in northern Ghana are crop-related, soil-related or involve cultural practices. Livestock-based agricultural adaptations include indigenous strategies such as adjusting quantities of feed to feed livestock, storing enough feed during the abundant period to be fed to livestock during the lean season, treating wounds with solution of certain barks of trees, and keeping local breeds which are already adapted to the climate of northern Ghana; and livestock production technologies to include breeding, health, feed/nutrition and housing.

The choice and adoption of adaptation strategies is variously contingent on demographic factors such as the household size, age, gender and education of the household head; economic factors such as income source; farm size; knowledge of adaptation options; and expectation of future prospects.

Eastern Africa adaptation measures

In Eastern Africa adaptation options are varied, including improving use of climate information, actions in the agriculture and livestock sector, and in the water sector.

Making better use of climate and weather data, weather forecasts, and other management tools enables timely information and preparedness of people in the sectors such as agriculture that depend on weather outcomes. This means mastering hydro-meteorological information and early warning systems. It has been argued that the indigenous communities possess knowledge on historical climate changes through environmental signs (e.g. appearance and migration of certain birds, butterflies etc.), and thus promoting of indigenous knowledge has been considered an important adaptation strategy.

Adaptation in the agricultural sector includes increased use of manure and crop-specific fertilizer, use of resistant varieties of crops and early maturing crops. Manure, and especially animal manure is thought to retain water and have essential microbes that breakdown nutrients making them available to plants, as compared to synthetic fertilizers that have compounds which when released to the environment due to over-use release greenhouse gases. One major vulnerability of the agriculture sector in Eastern Africa is the dependence on rain-fed agriculture. An adaptation solution is efficient irrigation mechanisms and efficient water storage and use. Drip irrigation has especially been identified as a water-efficient option as it directs the water to the root of the plant with minimal wastage. Countries like Rwanda and Kenya have prioritized developing irrigated areas by gravity water systems from perennial streams and rivers in zones vulnerable to prolonged droughts. During heavy rains, many areas experience flooding resulting from bare grounds due to deforestation and little land cover. Adaptation strategies proposed for this is promoting conservation efforts on land protection, by planting indigenous trees, protecting water catchment areas and managing grazing lands through zoning.

For the livestock sector, adaptation options include managing production through sustainable land and pasture management in the ecosystems. This includes promoting hay and fodder production methods e.g. through irrigation and use of waste treated water, and focusing on investing in hay storage for use during dry seasons. Keeping livestock is considered a livelihood rather than an economic activity. Throughout Eastern Africa Countries especially in the ASALs regions, it is argued that promoting commercialisation of livestock is an adaptation option. This involves adopting economic models in livestock feed production, animal traceability, promoting demand for livestock products such as meat, milk and leather and linking to niche markets to enhance businesses and provide disposable income.

In the water sector, options include efficient use of water for households, animals and industrial consumption and protection of water sources. Campaigns such as planting indigenous trees in water catchment areas, controlling human activities near catchment areas especially farming and settlement have been carried out to help protect water resources and avail access to water for communities especially during climatic shocks.

Comoros – "NAPA is the operational extension of the Poverty Reduction Strategy Paper (PRSP), as it includes among its adaptation priorities, agriculture, fishing, water, housing, health, but also tourism, in an indirect way, through the reconstitution of basin slopes and the fight against soils erosion, and therefore the protection of reefs by limiting the silting up by terrigenous contributions."

Kenya gazetted the Climate Change Act, 2016 which establishes an authority to oversee development, management, implementation and regulation of mechanisms to enhance climate change resilience and low carbon development for sustainable development, by the National and County Governments, the private sector, civil society, and other actors. Kenya has also developed the National Climate Change Action Plan (NCCAP 2018–2022 Archived 23 December 2019 at the Wayback Machine) which aims to further the country's development goals by providing mechanisms and measures to achieve low carbon climate-resilient development in a manner that prioritizes adaptation.

Central Africa adaptation measures

Angola – "The objective of the National Adaptation Programs of Action are to identify and communicate the urgent and immediate needs of the country regarding climate change adaptation, to increase Angola's resilience to climate variabilities and to climate change to ensure achievement of Poverty reduction programs, sustainable development objectives and the Millennium Development Goals pursued by the Government."

Southern Africa adaptation measures

There have been several initiatives at local (site-specific), local, national and regional scales aimed at strengthening to climate change. Some of these are: The Regional Climate Change Programme (RCCP), SASSCAL, ASSAR, UNDP Climate Change Adaptation, RESILIM, FRACTAL.  South Africa implemented the Long-Term adaptation Scenarios Flagship Research Programme (LTAS) from April 2012 to June 2014. This research also produced factsheets and a technical report covering the SADC region entitled "Climate Change Adaptation: Perspectives for the Southern African Development Community (SADC)".

Madagascar – the priority sectors for adaptation are: agriculture and livestock, forestry, public health, water resources and coastal zones.

Malawi – The NAPA identifies the following as high priority activities for adaptation: "Improving community resilience to climate change through the development of sustainable rural livelihoods, Restoring forests in the Upper and Lower Shire Valleys catchments to reduce siltation and associated water flow problems, Improving agricultural production under erratic rains and changing climatic conditions, Improving Malawi's preparedness to cope with droughts and floods, and Improving climate monitoring to enhance Malawi's early warning capability and decision making and sustainable utilisation of Lake Malawi and lakeshore areas resources". And according to the World Bank's Country Climate and Development Report (CCDR) for Malawi, can "take steps to jumpstart investments in climate-resilient infrastructure and halt land degradation and forest loss to improve agriculture productivity and carbon capture."

Mauritius – adaptation should address the following priority areas: coastal resources, agriculture, water resources, fisheries, health and well-being, land use change and forestry and biodiversity.

Mozambique – "The proposed adaptation initiatives target various areas of economic and social development, and outline projects related to the reduction of impacts to natural disasters, the creation of adaptation measures to climate change, fight against soil erosion in areas of high desertification and coastal zones, reforestation and the management of water resources."

Rwanda has developed the National Adaptation Programme of Action (NAPA 2006) which contains information to guide national policy-makers and planners on priority vulnerabilities and adaptations in important economic sectors. The country has also developed sector based policies on adaptation to climate change such as the Vision 2020, the National Environmental Policy and the Agricultural Policy among others.

Tanzania – Tanzania has outlined priority adaptation measures in their NAPA, and various national sector strategies and research outputs. The NAPA has been successful at encouraging climate change mainstreaming into sector policies in Tanzania; however, the cross-sectoral collaboration crucial to implementing adaptation strategies remains limited due to institutional challenges such as power imbalances, budget constraints and an ingrained sectoral approach. Most of the projects in Tanzania concern agriculture and water resource management (irrigation, water saving, rainwater collection); however, energy and tourism also play an important role.

Zambia – "The NAPA identifies 39 urgent adaptation needs and 10 priority areas within the sectors of agriculture and food security (livestock, fisheries and crops), energy and water, human health, natural resources and wildlife."

Zimbabwe – "The other strategic interventions by the NAP process will be: Strengthening the role of private sector in adaptation planning, Enhancing of the capacity of Government to develop bankable projects through trainings, Improving management of background climate information to inform climate change planning, Crafting a proactive resource-mobilization strategy for identifying and applying for international climate finance as requests for funds are primarily reactive at present, focusing on emergency relief rather than climate change risk reduction, preparedness and adaptation, Developing a coordinated monitoring and evaluation policy for programs and projects, as many institutions within the government do not currently have a systematic approach to monitoring and evaluation."

Lesotho – "The key objectives of the NAPA process entail: identification of communities and livelihoods most vulnerable to climate change, generating a list of activities that would form a core of the national adaptation program of action, and to communicate the country's immediate and urgent needs and priorities for building capacity for adaptation to climate change."

Namibia – the critical themes for adaptation are "Food security and sustainable biological resource base, Sustainable water resources base, Human health and well being and Infrastructure development.

South Africa has adopted in August 2020 its National Climate Change Adaptation Strategy, which "acts as a common reference point for climate change adaptation efforts in South Africa, and it provides a platform upon which national climate change adaptation objectives for the country can be articulated so as to provide overarching guidance to all sectors of the economy."

Society and culture

Inequality in climate research

Even though Africa is going to be one of the most affected continents from climate change, systematic inequity and other biases related to scientific research and funding mean that very little of the published science about climate change and climate research funding is for African scientist. An analysis of research money from 1990 to 2020 for climate change, found that 78% of research money for research on climate change in Africa was spent in European and North American institutions and more was spent for former British colonies than other countries. This pattern of parachute science, in turn both prevents local researchers from doing groundbreaking work, because they do not have the funding for experimental activities and reduces investment in local researchers ideas and in topics important to the Global South, such as climate change adaptation.

Accurate sustainability evaluations are challenging due to a lack of sustainable investment frameworks, as well as data and managerial capability restrictions. Currently, fewer than half of Africa's top pension funds report information on sustainability policies and execution.

The United Nations Conference on Trade and Development - International Standards of Accounting and Reporting (UNCTAD-ISAR) founded the African Regional Partnership for Sustainability and SDG Reporting in 2022. The collaboration has 53 members as of March 2023, including national corporate social responsibility networks and/or ministries from 27 African nations.

Explainable artificial intelligence

Within artificial intelligence (AI), explainable AI (XAI), generally overlapping with interpretable AI or explainable machine learning (XML), is a field of research that explores methods that provide humans with the ability of intellectual oversight over AI algorithms. The main focus is on the reasoning behind the decisions or predictions made by the AI algorithms, to make them more understandable and transparent. This addresses users' requirement to assess safety and scrutinize the automated decision making in applications. XAI counters the "black box" tendency of machine learning, where even the AI's designers cannot explain why it arrived at a specific decision.

XAI hopes to help users of AI-powered systems perform more effectively by improving their understanding of how those systems reason. XAI may be an implementation of the social right to explanation. Even if there is no such legal right or regulatory requirement, XAI can improve the user experience of a product or service by helping end users trust that the AI is making good decisions. XAI aims to explain what has been done, what is being done, and what will be done next, and to unveil which information these actions are based on. This makes it possible to confirm existing knowledge, challenge existing knowledge, and generate new assumptions.

Background

Machine learning (ML) algorithms used in AI can be categorized as white-box or black-box. White-box models provide results that are understandable to experts in the domain. Black-box models, on the other hand, are extremely hard to explain and may not be understood even by domain experts. XAI algorithms follow the three principles of transparency, interpretability, and explainability.

  • A model is transparent "if the processes that extract model parameters from training data and generate labels from testing data can be described and motivated by the approach designer."
  • Interpretability describes the possibility of comprehending the ML model and presenting the underlying basis for decision-making in a way that is understandable to humans.
  • Explainability is a concept that is recognized as important, but a consensus definition is not yet available; one possibility is "the collection of features of the interpretable domain that have contributed, for a given example, to producing a decision (e.g., classification or regression)".

In summary, Interpretability refers to the user's ability to understand model outputs, while Model Transparency includes Simulatability (reproducibility of predictions), Decomposability (intuitive explanations for parameters), and Algorithmic Transparency (explaining how algorithms work). Model Functionality focuses on textual descriptions, visualization, and local explanations, which clarify specific outputs or instances rather than entire models. All these concepts aim to enhance the comprehensibility and usability of AI systems. If algorithms fulfill these principles, they provide a basis for justifying decisions, tracking them and thereby verifying them, improving the algorithms, and exploring new facts.

Sometimes it is also possible to achieve a high-accuracy result with white-box ML algorithms. These algorithms have an interpretable structure that can be used to explain predictions. Concept Bottleneck Models, which use concept-level abstractions to explain model reasoning, are examples of this and can be applied in both image and text prediction tasks. This is especially important in domains like medicine, defense, finance, and law, where it is crucial to understand decisions and build trust in the algorithms. Many researchers argue that, at least for supervised machine learning, the way forward is symbolic regression, where the algorithm searches the space of mathematical expressions to find the model that best fits a given dataset.

AI systems optimize behavior to satisfy a mathematically specified goal system chosen by the system designers, such as the command "maximize the accuracy of assessing how positive film reviews are in the test dataset." The AI may learn useful general rules from the test set, such as "reviews containing the word "horrible" are likely to be negative." However, it may also learn inappropriate rules, such as "reviews containing 'Daniel Day-Lewis' are usually positive"; such rules may be undesirable if they are likely to fail to generalize outside the training set, or if people consider the rule to be "cheating" or "unfair." A human can audit rules in an XAI to get an idea of how likely the system is to generalize to future real-world data outside the test set.

Goals

Cooperation between agents – in this case, algorithms and humans – depends on trust. If humans are to accept algorithmic prescriptions, they need to trust them. Incompleteness in formal trust criteria is a barrier to optimization. Transparency, interpretability, and explainability are intermediate goals on the road to these more comprehensive trust criteria. This is particularly relevant in medicine, especially with clinical decision support systems (CDSS), in which medical professionals should be able to understand how and why a machine-based decision was made in order to trust the decision and augment their decision-making process.

AI systems sometimes learn undesirable tricks that do an optimal job of satisfying explicit pre-programmed goals on the training data but do not reflect the more nuanced implicit desires of the human system designers or the full complexity of the domain data. For example, a 2017 system tasked with image recognition learned to "cheat" by looking for a copyright tag that happened to be associated with horse pictures rather than learning how to tell if a horse was actually pictured. In another 2017 system, a supervised learning AI tasked with grasping items in a virtual world learned to cheat by placing its manipulator between the object and the viewer in a way such that it falsely appeared to be grasping the object.

One transparency project, the DARPA XAI program, aims to produce "glass box" models that are explainable to a "human-in-the-loop" without greatly sacrificing AI performance. Human users of such a system can understand the AI's cognition (both in real-time and after the fact) and can determine whether to trust the AI. Other applications of XAI are knowledge extraction from black-box models and model comparisons. In the context of monitoring systems for ethical and socio-legal compliance, the term "glass box" is commonly used to refer to tools that track the inputs and outputs of the system in question, and provide value-based explanations for their behavior. These tools aim to ensure that the system operates in accordance with ethical and legal standards, and that its decision-making processes are transparent and accountable. The term "glass box" is often used in contrast to "black box" systems, which lack transparency and can be more difficult to monitor and regulate. The term is also used to name a voice assistant that produces counterfactual statements as explanations.

Explainability and interpretability techniques

There is a subtle difference between the terms explainability and interpretability in the context of AI.

Term Definition Source
Interpretability "level of understanding how the underlying (AI) technology works" ISO/IEC TR 29119-11:2020(en), 3.1.42
Explainability "level of understanding how the AI-based system ... came up with a given result" ISO/IEC TR 29119-11:2020(en), 3.1.31

Some explainability techniques don't involve understanding how the model works, and may work across various AI systems. Treating the model as a black box and analyzing how marginal changes to the inputs affect the result sometimes provides a sufficient explanation.

Explainability

Explainability is useful for ensuring that AI models are not making decisions based on irrelevant or otherwise unfair criteria. For classification and regression models, several popular techniques exist:

  • Partial dependency plots show the marginal effect of an input feature on the predicted outcome.
  • SHAP (SHapley Additive exPlanations) enables visualization of the contribution of each input feature to the output. It works by calculating Shapley values, which measure the average marginal contribution of a feature across all possible combinations of features.
  • Feature importance estimates how important a feature is for the model. It is usually done using permutation importance, which measures the performance decrease when it the feature value randomly shuffled across all samples.
  • LIME (Local Interpretable Model-Agnostic Explanations method) approximates locally a model's outputs with a simpler, interpretable model.
  • Multitask learning provides a large number of outputs in addition to the target classification. These other outputs can help developers deduce what the network has learned.

For images, saliency maps highlight the parts of an image that most influenced the result.

Systems that are expert or knowledge based are software systems that are made by experts. This system consists of a knowledge based encoding for the domain knowledge. This system is usually modeled as production rules, and someone uses this knowledge base which the user can question the system for knowledge. In expert systems, the language and explanations are understood with an explanation for the reasoning or a problem solving activity.

However, these techniques are not very suitable for language models like generative pretrained transformers. Since these models generate language, they can provide an explanation, but which may not be reliable. Other techniques include attention analysis (examining how the model focuses on different parts of the input), probing methods (testing what information is captured in the model's representations), causal tracing (tracing the flow of information through the model) and circuit discovery (identifying specific subnetworks responsible for certain behaviors). Explainability research in this area overlaps significantly with interpretability and alignment research.

Interpretability

Grokking is an example of phenomenon studied in interpretability. It involves a model that initially memorizes all the answers (overfitting), but later adopts an algorithm that generalizes to unseen data.

Scholars sometimes use the term "mechanistic interpretability" to refer to the process of reverse-engineering artificial neural networks to understand their internal decision-making mechanisms and components, similar to how one might analyze a complex machine or computer program.

Studying the interpretability of the most advanced foundation models often involves searching for an automated way to identify "features" in generative pretrained transformers. In a neural network, a feature is a pattern of neuron activations that corresponds to a concept. A compute-intensive technique called "dictionary learning" makes it possible to identify features to some degree. Enhancing the ability to identify and edit features is expected to significantly improve the safety of frontier AI models.

For convolutional neural networks, DeepDream can generate images that strongly activate a particular neuron, providing a visual hint about what the neuron is trained to identify.

Knowledge localization

Large language models (LLMs), such as transformer-based models (GPT), possess the ability to engage in conversation using general knowledge. This capability raises the question of how, exactly, such knowledge is stored within the model.

Research has suggested that the model’s MLP component (the feed-forward layers) is the main site in which knowledge is stored, encoding information through associative links that function as key–value memories: each key corresponds to textual patterns in the training data, while each value induces a distribution over the output vocabulary.

A 2022 study aimed at locating where knowledge resides in the model employed a technique known as Causal Tracing. In tasks requiring general knowledge, the researchers injected noise into the hidden activations of the model, preventing it from completing the task. They then restored the clean activation values (taken from a noise-free run) to a different part of the model each time, observing when the model regained its ability to produce the correct answer. Based on these results, the authors concluded that factual knowledge is stored primarily in the MLP components of the model’s middle layers. They further proposed that model editing would be most effective in those regions, though this claim was later called into question.

Later studies suggest that, in most cases, factual information is distributed across the model rather than localized within a single layer. According to one version of this view, different layers encode different aspects of the same association. For example, a question about the capital of Japan may activate representations related to “Japan” in one layer and representations corresponding to “capital cities” in another; the combination of these representations yields the concept of “Tokyo.”

History and methods

During the 1970s to 1990s, symbolic reasoning systems, such as MYCIN, GUIDON, SOPHIE, and PROTOS could represent, reason about, and explain their reasoning for diagnostic, instructional, or machine-learning (explanation-based learning) purposes. MYCIN, developed in the early 1970s as a research prototype for diagnosing bacteremia infections of the bloodstream, could explain which of its hand-coded rules contributed to a diagnosis in a specific case. Research in intelligent tutoring systems resulted in developing systems such as SOPHIE that could act as an "articulate expert", explaining problem-solving strategy at a level the student could understand, so they would know what action to take next. For instance, SOPHIE could explain the qualitative reasoning behind its electronics troubleshooting, even though it ultimately relied on the SPICE circuit simulator. Similarly, GUIDON added tutorial rules to supplement MYCIN's domain-level rules so it could explain the strategy for medical diagnosis. Symbolic approaches to machine learning relying on explanation-based learning, such as PROTOS, made use of explicit representations of explanations expressed in a dedicated explanation language, both to explain their actions and to acquire new knowledge.

In the 1980s through the early 1990s, truth maintenance systems (TMS) extended the capabilities of causal-reasoning, rule-based, and logic-based inference systems. A TMS explicitly tracks alternate lines of reasoning, justifications for conclusions, and lines of reasoning that lead to contradictions, allowing future reasoning to avoid these dead ends. To provide an explanation, they trace reasoning from conclusions to assumptions through rule operations or logical inferences, allowing explanations to be generated from the reasoning traces. As an example, consider a rule-based problem solver with just a few rules about Socrates that concludes he has died from poison:

By just tracing through the dependency structure the problem solver can construct the following explanation: "Socrates died because he was mortal and drank poison, and all mortals die when they drink poison. Socrates was mortal because he was a man and all men are mortal. Socrates drank poison because he held dissident beliefs, the government was conservative, and those holding conservative dissident beliefs under conservative governments must drink poison."

By the 1990s researchers began studying whether it is possible to meaningfully extract the non-hand-coded rules being generated by opaque trained neural networks. Researchers in clinical expert systems who created neural network-powered decision support for clinicians sought to develop dynamic explanations that allow these technologies to be more trusted and trustworthy in practice. In the 2010s public concerns about racial and other bias in the use of AI for criminal sentencing decisions and findings of creditworthiness may have led to increased demand for transparent artificial intelligence. As a result, many academics and organizations are developing tools to help detect bias in their systems.

Marvin Minsky et al. raised the issue that AI can function as a form of surveillance, with the biases inherent in surveillance, suggesting HI (Humanistic Intelligence) as a way to create a more fair and balanced "human-in-the-loop" AI.

Explainable AI has been recently a new topic researched amongst the context of modern deep learning. Modern complex AI techniques, such as deep learning, are naturally opaque. To address this issue, methods have been developed to make new models more explainable and interpretable. This includes layerwise relevance propagation (LRP), a technique for determining which features in a particular input vector contribute most strongly to a neural network's output, although this technique has been shown to suffer from several important issues. Other techniques explain some particular prediction made by a (nonlinear) black-box model, a goal referred to as "local interpretability". We still today cannot explain the output of today's DNNs without the new explanatory mechanisms, we also can't by the neural network, or external explanatory components  There is also research on whether the concepts of local interpretability can be applied to a remote context, where a model is operated by a third-party.

There has been work on making glass-box models which are more transparent to inspection. This includes decision treesBayesian networks, sparse linear models, and more. The Association for Computing Machinery Conference on Fairness, Accountability, and Transparency (ACM FAccT) was established in 2018 to study transparency and explainability in the context of socio-technical systems, many of which include artificial intelligence.

Some techniques allow visualisations of the inputs to which individual software neurons respond to most strongly. Several groups found that neurons can be aggregated into circuits that perform human-comprehensible functions, some of which reliably arise across different networks trained independently.

There are various techniques to extract compressed representations of the features of given inputs, which can then be analysed by standard clustering techniques. Alternatively, networks can be trained to output linguistic explanations of their behaviour, which are then directly human-interpretable. Model behaviour can also be explained with reference to training data—for example, by evaluating which training inputs influenced a given behaviour the most, or by approximating its predictions using the most similar instances from the training data.

The use of explainable artificial intelligence (XAI) in pain research, specifically in understanding the role of electrodermal activity for automated pain recognition: hand-crafted features and deep learning models in pain recognition, highlighting the insights that simple hand-crafted features can yield comparative performances to deep learning models and that both traditional feature engineering and deep feature learning approaches rely on simple characteristics of the input time-series data.

Regulation

As regulators, official bodies, and general users come to depend on AI-based dynamic systems, clearer accountability will be required for automated decision-making processes to ensure trust and transparency. The first global conference exclusively dedicated to this emerging discipline was the 2017 International Joint Conference on Artificial Intelligence: Workshop on Explainable Artificial Intelligence (XAI). It has evolved over the years, with various workshops organised and co-located to many other international conferences, and it has now a dedicated global event, "The world conference on eXplainable Artificial Intelligence", with its own proceedings.

The European Union introduced a right to explanation in the General Data Protection Regulation (GDPR) to address potential problems stemming from the rising importance of algorithms. The implementation of the regulation began in 2018. However, the right to explanation in GDPR covers only the local aspect of interpretability. In the United States, insurance companies are required to be able to explain their rate and coverage decisions. In France the Loi pour une République numérique (Digital Republic Act) grants subjects the right to request and receive information pertaining to the implementation of algorithms that process data about them.

Limitations

Despite ongoing endeavors to enhance the explainability of AI models, they persist with several inherent limitations.

Adversarial parties

By making an AI system more explainable, we also reveal more of its inner workings. For example, the explainability method of feature importance identifies features or variables that are most important in determining the model's output, while the influential samples method identifies the training samples that are most influential in determining the output, given a particular input. Adversarial parties could take advantage of this knowledge.

For example, competitor firms could replicate aspects of the original AI system in their own product, thus reducing competitive advantage. An explainable AI system is also susceptible to being "gamed"—influenced in a way that undermines its intended purpose. One study gives the example of a predictive policing system; in this case, those who could potentially "game" the system are the criminals subject to the system's decisions. In this study, developers of the system discussed the issue of criminal gangs looking to illegally obtain passports, and they expressed concerns that, if given an idea of what factors might trigger an alert in the passport application process, those gangs would be able to "send guinea pigs" to test those triggers, eventually finding a loophole that would allow them to "reliably get passports from under the noses of the authorities".

Adaptive integration and explanation

Many approaches that it uses provides explanation in general, it doesn't take account for the diverse backgrounds and knowledge level of the users. This leads to challenges with accurate comprehension for all users. Expert users can find the explanations lacking in depth, and are oversimplified, while a beginner user may struggle understanding the explanations as they are complex. This limitation downplays the ability of the XAI techniques to appeal to their users with different levels of knowledge, which can impact the trust from users and who uses it. The quality of explanations can be different amongst their users as they all have different expertise levels, including different situation and conditions.

Technical complexity

A fundamental barrier to making AI systems explainable is the technical complexity of such systems. End users often lack the coding knowledge required to understand software of any kind. Current methods used to explain AI are mainly technical ones, geared toward machine learning engineers for debugging purposes, rather than toward the end users who are ultimately affected by the system, causing "a gap between explainability in practice and the goal of transparency". Proposed solutions to address the issue of technical complexity include either promoting the coding education of the general public so technical explanations would be more accessible to end users, or providing explanations in layperson terms.

The solution must avoid oversimplification. It is important to strike a balance between accuracy – how faithfully the explanation reflects the process of the AI system – and explainability – how well end users understand the process. This is a difficult balance to strike, since the complexity of machine learning makes it difficult for even ML engineers to fully understand, let alone non-experts.

Understanding versus trust

The goal of explainability to end users of AI systems is to increase trust in the systems, even "address concerns about lack of 'fairness' and discriminatory effects". However, even with a good understanding of an AI system, end users may not necessarily trust the system. In one study, participants were presented with combinations of white-box and black-box explanations, and static and interactive explanations of AI systems. While these explanations served to increase both their self-reported and objective understanding, it had no impact on their level of trust, which remained skeptical.

This outcome was especially true for decisions that impacted the end user in a significant way, such as graduate school admissions. Participants judged algorithms to be too inflexible and unforgiving in comparison to human decision-makers; instead of rigidly adhering to a set of rules, humans are able to consider exceptional cases as well as appeals to their initial decision. For such decisions, explainability will not necessarily cause end users to accept the use of decision-making algorithms. We will need to either turn to another method to increase trust and acceptance of decision-making algorithms, or question the need to rely solely on AI for such impactful decisions in the first place.

However, some emphasize that the purpose of explainability of artificial intelligence is not to merely increase users' trust in the system's decisions, but to calibrate the users' level of trust to the correct level. According to this principle, too much or too little user trust in the AI system will harm the overall performance of the human-system unit. When the trust is excessive, the users are not critical of possible mistakes of the system and when the users do not have enough trust in the system, they will not exhaust the benefits inherent in it.

Criticism

Some scholars have suggested that explainability in AI should be considered a goal secondary to AI effectiveness, and that encouraging the exclusive development of XAI may limit the functionality of AI more broadly. Critiques of XAI rely on developed concepts of mechanistic and empiric reasoning from evidence-based medicine to suggest that AI technologies can be clinically validated even when their function cannot be understood by their operators.

Some researchers advocate the use of inherently interpretable machine learning models, rather than using post-hoc explanations in which a second model is created to explain the first. This is partly because post-hoc models increase the complexity in a decision pathway and partly because it is often unclear how faithfully a post-hoc explanation can mimic the computations of an entirely separate model. However, another view is that what is important is that the explanation accomplishes the given task at hand, and whether it is pre or post-hoc doesn't matter. If a post-hoc explanation method helps a doctor diagnose cancer better, it is of secondary importance whether it is a correct/incorrect explanation.

The goals of XAI amount to a form of lossy compression that will become less effective as AI models grow in their number of parameters. Along with other factors this leads to a theoretical limit for explainability.

Explainability in social choice

Explainability was studied also in social choice theory. Social choice theory aims at finding solutions to social decision problems, that are based on well-established axioms. Ariel D. Procaccia explains that these axioms can be used to construct convincing explanations to the solutions. This principle has been used to construct explanations in various subfields of social choice.

Voting

Cailloux and Endriss present a method for explaining voting rules using the axioms that characterize them. They exemplify their method on the Borda voting rule .

Peters, Procaccia, Psomas and Zhou present an algorithm for explaining the outcomes of the Borda rule using O(m2) explanations, and prove that this is tight in the worst case.

Participatory budgeting

Yang, Hausladen, Peters, Pournaras, Fricker and Helbing present an empirical study of explainability in participatory budgeting. They compared the greedy and the equal shares rules, and three types of explanations: mechanism explanation (a general explanation of how the aggregation rule works given the voting input), individual explanation (explaining how many voters had at least one approved project, at least 10000 CHF in approved projects), and group explanation (explaining how the budget is distributed among the districts and topics). They compared the perceived trustworthiness and fairness of greedy and equal shares, before and after the explanations. They found out that, for MES, mechanism explanation yields the highest increase in perceived fairness and trustworthiness; the second-highest was Group explanation. For Greedy, Mechanism explanation increases perceived trustworthiness but not fairness, whereas Individual explanation increases both perceived fairness and trustworthiness. Group explanation decreases the perceived fairness and trustworthiness.

Payoff allocation

Nizri, Azaria and Hazon present an algorithm for computing explanations for the Shapley value. Given a coalitional game, their algorithm decomposes it to sub-games, for which it is easy to generate verbal explanations based on the axioms characterizing the Shapley value. The payoff allocation for each sub-game is perceived as fair, so the Shapley-based payoff allocation for the given game should seem fair as well. An experiment with 210 human subjects shows that, with their automatically generated explanations, subjects perceive Shapley-based payoff allocation as significantly fairer than with a general standard explanation.

Logical reasoning

From Wikipedia, the free encyclopedia https://en.wikipedia.org/wiki/Logical_reasoning   Logical reasoni...