AI For Real

Category: AI Research

  • Developing AI That Can Tell What You Mean Before You Even Mouth The Words

    Developing AI That Can Tell What You Mean Before You Even Mouth The Words

    Imagine telling an AI to do something and watching it make the wrong move.

    But before you can react with more words, the machine already knows something is off. It detects, from your brain activity, that its interpretation does not match what you intended and adjusts what it is doing! That may sound like AI can read your mind before you have expressed the thought.

    Korean researchers have developed an AI system that can use brain signals to detect when a machine has misunderstood a person’s intention.

    Well, it cannot read your thoughts in the science-fiction sense, but researchers at South Korea’s The Korea Advanced Institute of Science and Technology (KAIST) have taken an important step towards AI detecting when its understanding of your intention is wrong.

    The technology, developed by a KAIST team led by Professor Sang Wan Lee in collaboration with Microsoft Research Asia, is called Neural Value Alignment, or NVA. The basic idea is surprisingly simple: instead of making AI rely only on what people say or write, the researchers want it to also detect signals from the brain that reveal when something has gone wrong.

    Consider a familiar situation. You ask an AI assistant or robot to perform a task. It follows your instruction, but does something you did not actually intend. Normally, you have to correct it by saying something such as, “No, that’s not what I meant.” KAIST’s system is designed to detect that mismatch from brain activity, potentially allowing the machine to correct itself without waiting for the person to explain.

    This is possible because the brain reacts when reality does not match expectations. The researchers used electroencephalography, or EEG, to record electrical activity from people’s brains while they interacted with an AI system.

    They focused on two types of signals. One indicates that the result was different from what a person expected. The other indicates that the situation itself has not developed as the person expected. The technical names are reward prediction error and state prediction error, but the important point is that both can provide clues about what a person thinks should be happening.

    The researchers found that these signals could be distinguished in brain activity. They then used them together to help an AI work out two separate things: whether it had chosen the wrong action, and whether it had misunderstood the person’s goal in the first place. That distinction matters because the same action can sometimes serve different purposes, while different actions can achieve the same goal.

    The research is still a long way from an AI that casually reads people’s minds. The experiment relies on EEG equipment to measure brain signals, and the researchers are demonstrating a scientific method for detecting mismatches between human intentions and AI behaviour. It is not a consumer product that can currently be plugged into ChatGPT, a phone or a home robot.

    The next stage is to develop this idea for more natural cooperation between people and machines. KAIST says the approach could eventually be applied to physical robots in homes and factories, autonomous vehicles, medical robots and rehabilitation systems. The team’s ongoing work with Microsoft Research Asia is aimed at developing brain-computer technologies that allow humans and AI to communicate and collaborate more naturally.

    There is no announcement that this particular technology is being released as a commercial product. For now, it remains a research advance. But its significance is clear: the future of AI may not depend only on machines understanding what we say. It may also depend on machines becoming better at recognising when what they have understood is not what we actually meant.

    Reference

    https://pubmed.ncbi.nlm.nih.gov/42636135

  • When Thinking Harder Can Make It Worse For An AI Model

    When Thinking Harder Can Make It Worse For An AI Model

    For much of the past two years, one of the clearest ideas in frontier AI has been that reasoning models get better when they are given more time to think.

    The recipe sounds intuitive. Instead of asking a model to jump straight to an answer, let it generate a long chain of intermediate reasoning. Give it more test-time compute, and difficult problems become easier. In effect, the model gets to spend more time working things out before committing to an answer.

    But research emerging in 2026 is putting an important qualification on that idea: sometimes the model has already solved the problem, and then talks itself out of the correct answer.

    This phenomenon is increasingly being called “overthinking”.

    Why an LLM Overthinks

    A Google DeepMind study published in July examined the thought processes of open-source reasoning models including Qwen3 and distilled versions of DeepSeek-R1.

    Rather than simply counting how many tokens a model uses, the researchers broke its reasoning into smaller sub-thoughts and mapped how those thoughts connect. They identified patterns they call the “Explorer” and the “Late Landing”. Both point to over-exploration and over-verification as important drivers of unnecessary reasoning.

  • The 24 AI Risks Experts Are Watching Most Closely

    The 24 AI Risks Experts Are Watching Most Closely

    Artificial intelligence (AI) keeps getting more powerful, but which risks actually deserve our attention over the next few years?

    A team from MIT FutureTech and the University of Queensland set out to answer that. They asked 272 international AI experts to rate 24 AI risks by how likely they are and how much harm they could cause, using a structured survey method designed to build consensus across many rounds.

    The headline finding: even in a best-case scenario where companies and governments make reasonable efforts to manage AI responsibly, five risk categories still carry at least a 10% chance of catastrophic harm within five years — meaning outcomes on the scale of over a million deaths, $100 billion in losses, or comparable damage to society.

    Those five are AI systems gaining dangerous capabilities

    • AI-enabled weapons and cyberattacks
    • environmental harm
    • inequality and job loss
    • the concentration of power and unequal distribution of AI’s benefits.

    Without any mitigation at all, the picture is worse: 8 of the 24 risk domains crossed that same 10% catastrophic-probability threshold.

    Two risks stood out to researchers as especially urgent. AI is unusually good at coding and pattern recognition, which makes it a natural fit for accelerating cyberattacks. And “dangerous capabilities” is a broader worry — the same advances that help with legitimate work can also make surveillance, deepfakes, persuasion, and even weapons development easier to pull off.

    The study also flagged competitive pressure as a risk multiplier: when companies or countries fear falling behind, they may cut corners on safety just to move faster.

    Three sectors were named as most exposed: information (misinformation, privacy), national security (cyberattacks, weapons), and finance (fraud, market manipulation). And there’s a mismatch worth noting — developers and regulators are seen as most responsible for managing these risks, while everyday users bear the brunt of them.

    Reference:

    https://mitsloan.mit.edu/ideas-made-to-matter/these-are-most-urgent-ai-risks-according-to-272-experts

  • AI Tools May Be Eroding Professionals’ Skills, Researchers Warn

    AI Tools May Be Eroding Professionals’ Skills, Researchers Warn

    Artificial intelligence (AI) tools are helping professionals work faster, but growing evidence suggests they could also weaken the very skills people have spent years developing, prompting researchers to call for urgent studies on how to prevent AI driven deskilling.

    The concern is particularly strong in medicine. A survey of US health care workers published this month in nature found that 70% of nurses and 77% of physicians worry that over reliance on AI systems could reduce their clinical skills.

    Researchers say those fears are beginning to find support in real world evidence. The study published last October in The Lancet Gastroenterology and Hepatology found that experienced physicians became less effective at detecting precancerous intestinal growths when an AI tool they had grown accustomed to using was suddenly unavailable.

    The study followed physicians in Poland who specialized in endoscopy and had each performed at least 2,000 colonoscopies during their careers. Researchers introduced an AI system that analysed colonoscopy images in real time and highlighted adenomas, a type of precancerous lesion. The system was available only on certain days, allowing researchers to compare doctors’ performance with and without AI assistance.

    Before the technology was introduced, doctors detected at least one adenoma in 28.4% of colonoscopies. After three months of using the AI system, the detection rate during procedures performed without AI assistance fell to 22.4%.

    The findings suggest that even highly experienced clinicians may become less effective at tasks they routinely perform if they become dependent on AI, researchers said. The study authors argued that continuous exposure to AI tools could leave clinicians less motivated, less focused and less responsible when making decisions without technological support.

    The issue is not limited to medicine. Researchers at AI company Anthropic have also begun investigating whether software engineers lose skills when they rely heavily on AI coding assistants.

    In a randomized controlled trial involving 52 software engineers, participants completed a basic coding task with access to online resources, while half were also encouraged to use an AI assistant. The study is part of a broader effort to understand how AI affects long term expertise.

    As AI becomes a routine part of workplaces across industries, researchers say the challenge will be finding ways to capture the technology’s productivity benefits without allowing essential human expertise to fade.

  • Older Adults Are More Open To AI Than Many Assume, Global EY Study Finds

    Older Adults Are More Open To AI Than Many Assume, Global EY Study Finds

    Older generations are embracing artificial intelligence (AI) more readily than common stereotypes suggest, according to new global research from EY, challenging assumptions that people aged 60 and above are resistant to emerging technologies.

    The report, conducted by EY Ripples in collaboration with Microsoft, Kite Insights, OATS and OpenAI, surveyed 2,515 adults aged 60 to 85 across 16 countries. It found that while many older adults remain cautious about AI, a significant number are already using the technology for learning, health-related information and everyday tasks — and most report positive experiences.

    Only 24% of respondents described themselves as “quite” or “very familiar” with AI. However, researchers noted that familiarity does not necessarily reflect actual use, as many older adults interact with AI-powered tools embedded in search engines, banking applications and customer service platforms without realizing it.

    Usage patterns also varied considerably. Around two in five respondents said they had either never used AI or had only experimented with it once or twice. Conversely, approximately one in five reported using AI frequently, highlighting a growing divide within older populations themselves.

    Employment status emerged as a key factor influencing adoption. Older adults still in the workforce were three times more likely to use AI than those who had retired. Researchers suggested that continued workplace exposure gives employed individuals greater opportunities to build confidence with the technology.

    The survey also identified a gender gap in AI adoption. Nearly one-third of women surveyed said they had never used AI tools, compared with one in five men. The report linked this disparity to broader patterns in technology access and participation, including women’s lower representation in science and technology fields.

    Among those who do use AI, learning emerged as the most common application, followed by health and travel assistance. Participants generally reported positive experiences when using AI for work, education and creative activities.

    The findings suggest that older adults are not rejecting AI outright. Instead, many are approaching it with a combination of curiosity, pragmatism and caution — seeking clear guidance on how to use the technology safely and effectively in their everyday lives.

    Click here to read the report.

  • AI Poses A “Hidden Threat” To Organizations: Report

    AI Poses A “Hidden Threat” To Organizations: Report

    In December 2025, I had written in one of my newsletters:

    ….the core question is whether delegating cognitive tasks to AI makes us more efficient or quietly atrophies the mental skills we stop exercising. Neuroimaging research shows that people who used an LLM to complete cognitive tasks showed measurably less brain activity in networks associated with deep processing and critical thinking than those who worked independently.

    Researchers who used AI to compile papers were often unable to recall key information from their own work — cited as a telling sign of cognitive disengagement. Artificial intelligence (AI) is widely promoted as a force multiplier for organizations, promising efficiency, speed, and data-driven insights. Yet, a new analysis warns that careless adoption of AI risks eroding the very human skills and organizational structures that underpin long-term competitiveness.

    Now, a new article in Harvard Business Review (HBR) raises almost the same concerns, but for organizations implementing AI.

    The Core Concern

    • AI’s fluency and confidence create an illusion of competence, encouraging employees to offload critical thinking to machines, says the HBR article.
    • Over-reliance on AI can hollow out tacit knowledge, judgment, and interpretive reasoning—capabilities essential for innovation, crisis response, and strategic planning.
    • Organizations risk becoming technologically advanced but competitively fragile if they fail to protect human expertise.

    Three Ways AI Erodes Capabilities

    1. People Stop Thinking
      • Employees defer to AI outputs instead of developing their own analyses or strategies.
      • Example: Creston Telecom (Australia) found managers presenting AI-generated scenarios without being able to defend choices.
      • Solution: Instituted AI-free strategy sessions and a six-month “strategy residency” to preserve judgment and systems thinking.
    2. Rules Get Buried in Systems
      • AI embeds subjective, moral decisions (e.g., credit approvals, promotions) into opaque algorithms.
      • This undermines deliberation, accountability, and adaptability.
      • Example: Piedmont Regional Bank (U.S.) noticed its Credit Committee leaning heavily on AI.
        • Response: Quarterly “credit standards roundtables” to debate evolving criteria.
        • Introduced apprenticeships pairing junior analysts with senior lenders, ensuring judgment and accountability remain human-driven.
    3. Social Ties Are Weakened
      • AI displaces collaborative problem-solving, reducing trust and shared purpose.
      • Example: Brightview Creative (U.K. advertising agency) saw clients leaving despite strong campaign metrics.
        • Clients felt they were dealing with a “vending machine” rather than creative partners.
        • Solution: Banned AI-generated content in client presentations, appointed strategic leads to articulate human judgment, and rebuilt client confidence.

    Key Takeaway

    AI can enhance organizational performance—but it cannot:

    • Develop expertise through lived experience
    • Take moral responsibility
    • Build trust, courage, or shared purpose

    The report says these remain irreducibly human functions. Leaders must ensure AI augments rather than replaces them, or risk losing the competitive edge that makes their organizations resilient.

    Source: HBR

    What’s your view on the above? Do comment.

  • AI Is Writing Your Code, But Is It Rewriting Your Brain?

    AI Is Writing Your Code, But Is It Rewriting Your Brain?

    In early 2026, researchers at Anthropic published new findings on a question that’s quietly reshaping computer science education: How does AI assistance affect human coding skill formation?

    For experienced developers and students just getting started, the answer matters more than ever.

    AI coding assistants can autocomplete functions, generate boilerplate, explain stack traces, and even architect small applications from a prompt. Used well, they feel like a senior engineer looking over your shoulder. But Anthropic’s research suggests the impact on skill development depends heavily on how these tools are used.