AI making a difference to real people

Robot helping growthRay Poynter, 25 August 2025

I think too much of the discussion about AI is focused on what it might deliver in the future. I believe it is important that we look at what has already been achieved to get a sense of where we are and where it might go soon. This post presents a set of examples from around the world and across various fields.

Overview
This note highlights where artificial intelligence (AI) has delivered tangible results for people over the last 18 months. It not only focuses on good outcomes, but also on tangible impacts, some of which might be considered bad outcomes, such as job losses. The note examines various issues, including cost and headcount impacts in companies, frontline health in Africa, farming in China, community action in Latin America, education, science/pharma, and industry.

Companies: reducing costs and, in some cases, headcount
I think most of us would consider cost reductions as a good change, but cuts in headcount are a more contentious issue. As I stated above, I am showing evidence of change, not just evidence of good things.

  • Klarna’s AI assistant now handles around two‑thirds of customer chats, cutting resolution times from about 11 minutes to about 2 minutes. Management reported that AI helped shrink headcount primarily through attrition (from roughly 5,000 to 3,800) while lifting productivity.
  • IBM reports that AI now handles the majority of routine HR tasks. In 2025, the company said it had replaced hundreds of HR roles with AI and reinvested savings to hire engineers and sales staff. Leadership also linked AI to multi‑billion‑dollar productivity gains.

Healthcare in Africa: more cases found, earlier and closer to communities

  • Tuberculosis (TB): At the 2024 Union World Conference on Lung Health, teams from Ethiopia, Nigeria, and Kenya presented outcomes from community TB screening using AI-aided chest X-ray (CAD). Examples included 12,181 people screened in Ethiopia (AI sensitivity about 94%), nearly 26,000 scan reviews in Nigeria identifying otherwise missed cases, and 15,916 people screened in Kenya with high positivity among those with elevated CAD scores.
  • Uganda is scaling digital X‑ray with CAD through the Ministry of Health (48 portable units procured in 2025 to boost active case finding).
  • Maternal health: A 2024 pilot in Uganda uses AI ultrasound to help midwives date pregnancies accurately in clinics and outreach, encouraging earlier antenatal attendance and better risk management for complications (e.g., pre‑eclampsia).

Farming in China: fully unmanned wheat–maize rotation at scale

  • In Henan (Qingfeng county), a 233‑hectare fully unmanned farm reported faster harvesting (7 days down to 4), large reductions in labour for irrigation/fertiliser management (around 80–90%), about 40% lower human costs, and roughly 20% higher wheat yields versus conventional nearby fields, coordinated by AI with BeiDou‑guided machines and smart irrigation.

Communities & environment in Latin America: measurable water and forest outcomes

  • Water and smallholders: Kilimo (Argentina-born ag-tech) uses AI-guided irrigation to cut water use and share audited ‘volumetric water benefits’ with farmers; coverage spans thousands of farms across seven countries. Press reports in 2024 highlighted tens of millions of cubic metres saved and the typical net earnings uplift for participating farmers after revenue‑sharing for saved water was 20 to 40%.
  • Forest monitoring for enforcement: MapBiomas Alerta (a civil‑society/academic network using AI/ML with satellite data) underpins deforestation alerts used by authorities and NGOs across Brazil. MapBiomas reported a 32.4% fall in total deforestation in 2024 versus 2023 (context: enforcement, climate and fire dynamics also matter).

Education: controlled trials show sizable learning gains at low cost

  • Harvard RCT (2024–2025): In an undergraduate physics course, a randomised crossover study found students learned roughly twice as much, in less time, with a tailored AI tutor compared with instructor‑led active learning sessions. A preprint and later versions (2025) document the results.
  • Live tutoring at scale (2024): Tutor CoPilot (a human‑AI assistant for K‑12 tutors) was tested in an randomised control test with 900 tutors and 1,800 students; access to the tool raised student topic‑mastery by about four percentage points on average (and about nine percentage points for lower‑rated tutors) at around $20 per tutor per year.

Science & pharma: new capabilities and early translational outputs

  • Antibiotics designed with generative AI (2025): MIT researchers reported AI‑designed candidates active against drug‑resistant MRSA and N. gonorrhoeae, including efficacy in mouse models, published in Cell; the related team is progressing pre‑clinical optimisation with a non‑profit partner.
  • AlphaFold 3 (2024): DeepMind/Isomorphic Labs published a Nature paper showing markedly better accuracy for predicting protein–ligand and other biomolecular interactions, with an online server enabling academic access; debate continues on practical limits and validation, but the tool is already helping design work.
  • Drug pipelines: while no AI‑designed drug has yet reached market, multiple candidates advanced in the last 18 months, and industry consolidation continued (e.g., Exscientia joining Recursion).

Industry & manufacturing: fewer defects and faster engineering

  • A major automotive manufacturer scaled AI vision inspections (tens of millions of inspections) and reported immediate, significant cuts in defects and downtime (IBM case study, Feb 2024).
  • Industrial copilots: Siemens and partners expanded generative‑AI assistants on the shop‑floor in 2024–2025; pilots reported faster code generation and setup, easing skilled‑labour bottlenecks.

What this adds up to

Across sectors, the clearest near‑term wins are: (1) customer‑service automation (speed, quality and cost), (2) targeted screening and triage in health (higher case‑finding and earlier care), (3) resource efficiency in farming and industry (less water, fewer defects), and (4) learning support (large, cost‑effective gains in controlled studies). In pharma, the most significant impact is upstream: designing and filtering better candidates faster; clinical proof remains the gating step.

This ties in well with my experience of AI in market research. Teams leveraging LLMs are reporting to me that they are 15% to 25% more productive. Additionally, conversational data collection is making inroads, and most of the big players have some tools to leverage AI.

Notes (sources, all accessed Aug 2025)

Notes on how this article was created
My first step in creating this post was to use ChatGPT 5 and use Deep Research to make the initial draft, specifically asking it to list the sources in the notes.

Next, I uploaded this output to Gemini 2.5 Pro and asked it to check all of the sources.

Finally, I read and tweaked the text to put it in my voice and to ensure I was happy to own the story and argument.

This took about 30 minutes of computer time and maybe 45 minutes of my time. This is a lot faster than if I had used Google to search for references, read the references and created the initial draft that way. Fifteen of those minutes were spent checking the links and replacing two that I was not happy with.

In this case, I was looking for examples of AI in use. I was not looking for the ‘best examples’ or ‘all of the examples’. If I had wanted to find the best examples or all of the examples, the requirements would have been much more challenging to achieve and more difficult to check. This use is an example of utilising AI as an assistant. The key to the article (its strengths or its weaknesses) is a product of the prompting and framing I created. The edits I made to the initial draft make it smoother, but these edits were only practical because the initial draft was close enough to iterate from.

Leave a Reply

Your email address will not be published. Required fields are marked *