Bill Gates said AI’s recent progress is “quite stunning,” but argued that technological breakthroughs will not automatically benefit poorer countries. During a fireside chat with Infocomm Media Development Authority chairman Russell Tham at Singapore’s ATxInspire event, Gates described potential applications in healthcare, agriculture and education while warning that philanthropy, public policy and international cooperation may be needed to prevent AI from widening existing inequalities. Computer Weekly reported the remarks on May 7, 2025.
What Bill Gates said in Singapore
Gates’s comments came during a conversation with Russell Tham at ATxInspire in Singapore. As reported by Computer Weekly, Gates said he uses AI research tools several times a day and described the pace of progress as “quite stunning.” He also called for cooperation at a time of rapid technological change, political polarisation and geopolitical fragmentation.
The event was held in 2025, so these remarks should not be presented as Gates’s latest position in 2026. His argument was not that AI has already solved poverty or healthcare access. It was that falling technology costs could create an unusually powerful opportunity—provided poorer countries receive the investment and infrastructure needed to use it.
Why he thinks AI has crossed a threshold
Gates connected his enthusiasm to a long-standing challenge involving biology knowledge. He said he had wondered whether a computer could absorb enough material from a biology textbook to outperform a student taking an Advanced Placement biology examination.
#1 Best Overall
According to Gates’s account, that remained unsolved until GPT-4. He recalled challenging OpenAI to demonstrate that an AI system could pass the exam, and said the company showed him a system within eight months. In the demonstration he described, the system answered 19 of approximately 20 questions correctly; the missed question involved mathematics and the system’s then-current single-pass approach.
Those details are Gates’s recollection as reported by Computer Weekly, not an independently documented benchmark. The report does not provide the exam paper, test protocol, model version, demonstration date or a reproducible evaluation. The anecdote nevertheless illustrates why Gates views recent systems as qualitatively different from earlier software.
His forecast: intelligence could become much cheaper
Gates compared AI with the historical decline in the cost of computing. He forecast that many forms of intelligence—and eventually physical capabilities supplied by robots—could become largely free over the following decade.
“Free” in this context is best understood as a possible fall in the marginal cost of accessing an AI service, not the disappearance of all costs. Chips, electricity, mobile networks, devices, data collection, maintenance, cybersecurity, skilled oversight and regulation will still require money. A low-cost model is not the same as a low-cost service that works reliably in a rural clinic or school.
Where AI could help lower-income countries
Health information and clinical support
Gates envisaged AI providing health information or doctor-like guidance to people who rarely see a doctor. In practical terms, that could include symptom education, triage, translation, appointment support and tools that help frontline health workers.
Rank #2
That vision is narrower than giving everyone a doctor. AI cannot by itself provide physical examination, diagnostic tests, emergency treatment, medicines, hospitals or accountability for a clinical decision. Incorrect advice, missed emergencies, bias in training data, privacy breaches and unclear liability could cause serious harm. Any health deployment would need clinical validation, local-language testing, human escalation and regulation.
Advice for farmers
An AI adviser could help farmers interpret weather information, identify pests, compare crops or find market information. The value would depend on current local data and recommendations that reflect the farmer’s soil, climate, seed supply, finances and farming practices.
Generic advice can be dangerous when conditions differ from the model’s assumptions. Out-of-date weather or market data, poor language support and recommendations for unavailable inputs could turn an apparently useful tool into a source of crop losses. Agricultural extension workers and local experts would remain important.
AI tutors
AI tutors could explain concepts, provide practice exercises and adapt lessons to an individual student. They could extend support where teacher time is limited, including through inexpensive devices.
They would not replace the wider education system. Schools still need teachers, safe learning environments, curriculum oversight, assessment standards, child protection and access to electricity and connectivity. An effective program would integrate tutors with teachers and education ministries rather than treat software as a substitute for them.
Why equal access will not happen automatically
Gates’s central warning is economic: commercial incentives may not be strong enough to build the datasets, products and support systems that low-income markets need. Access depends on more than the price of an AI query.
- Connectivity and power: weak mobile coverage, unreliable electricity and expensive data can block use.
- Devices: feature phones could become delivery channels for limited services, but a full medical or tutoring experience may require more capable hardware.
- Language and culture: systems need local-language data, speech recognition and culturally appropriate interfaces.
- Data quality: health, farming and education tools require representative, current local information.
- Institutions: ministries, clinics and schools need procurement, training, monitoring and technical support.
- Trust and ownership: users and governments need clarity about who controls data, models and decisions.
These barriers create a risk that high-income countries capture most of the productivity gains while poorer countries remain dependent on imported systems.
Recommended Free Tools
Why Gates says philanthropy has a role
Gates argued that large technology companies may not prioritise work serving markets with limited purchasing power. Philanthropy can fund local-language datasets, model testing, pilot programs and deployment where the social return is high but the commercial return is uncertain.
It is not a complete solution. Governments, development agencies, universities, local companies, telecom operators and health and education systems would also need to contribute. Long-term public procurement and local technical capacity matter because a grant-funded pilot can fail when the grant ends.
The practical questions include who validates advice, who owns the resulting data, who pays for maintenance and who is liable when an automated recommendation causes harm. Those are governance and financing questions as much as engineering questions.
Drug discovery shows the wider implementation problem
Gates said AI is good at understanding chemical and protein shapes, but warned that regulation could become the bottleneck. If a drug can be discovered in six months while approval still takes five years, faster discovery alone will not transform treatment timelines. He said the Gates Foundation had funded work applying AI to drug safety and regulatory processes.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitchesThe example separates several stages that are often collapsed into the phrase “AI will speed medicine”:
- Discovering a candidate compound.
- Testing safety and efficacy.
- Obtaining regulatory approval.
- Manufacturing consistently at scale.
- Paying for treatment and securing reimbursement.
- Distributing it to patients.
AI may accelerate one stage while leaving the others unchanged. Gates’s point was about that system-wide mismatch, not evidence that drug development timelines have already fallen generally.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What smaller and open models could change
Gates praised the quality of open-source AI models and encouraged experimentation with smaller systems that use distillation to approach the performance of larger models. Smaller models can reduce computing costs, operate in places with limited connectivity and support on-device or locally hosted applications.
“Open-source” is not one guarantee. It may refer to model weights, source code, licensing or development practices, and those elements are not interchangeable. Open models can still have weak factual reliability, difficult licensing, security vulnerabilities, limited documentation and no dependable provider for updates or safety fixes. Hardware, electricity and local expertise remain costs even when the software is freely available.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Best Value
The possibility that AI widens the gap
The optimistic case assumes that cheaper intelligence spreads widely. The opposite outcome is also plausible: countries and companies that already control advanced chips, cloud capacity, data, capital and skilled workers may capture most of the gains.
That makes distribution a policy choice rather than an automatic consequence of technical progress. Lower-income countries need opportunities to adapt models to local priorities, develop their own expertise and negotiate fair data and procurement terms. Otherwise, they may receive tools designed elsewhere without meaningful control over how those tools affect public services.
Gates’s advice to AI entrepreneurs
Gates advised entrepreneurs to find a niche, be somewhat contrarian and pursue opportunities that others overlook. He linked that advice to areas such as malaria and malnutrition, where social need does not necessarily create intense commercial competition.
For businesses, that points to a gap between lucrative enterprise automation and public-interest applications. Building for underserved markets may require partnerships with governments and charities, patient financing and careful attention to local operating conditions. It can be commercially viable in some cases, but it should not be confused with a quick consumer software opportunity.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →The test for Gates’s vision
Gates’s message combines strong optimism about AI capabilities with a warning about economics and institutions. Healthcare guidance, farming advice and tutoring could extend access to useful information, but they are not substitutes for doctors, agronomists, teachers, hospitals, schools or functioning public systems.
Whether AI narrows or widens global inequality will depend on who funds local data and deployment, who controls the technology, how safety is enforced and whether infrastructure and skills reach users who cannot pay wealthy-market prices. The “stunning” progress is a technological claim; equal benefit is a financing, governance and political challenge.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
