Reading time: 4 minutes

At the latest Transylvania Lectures event on 24 April, organized by Mathias Corvinus Collegium (MCC), AI expert, computer scientist and author Dr. Emmanuel Maggiori spoke about the hidden realities, the drawbacks and the dangers of overconfidence of the AI industry. Despite its defects, millions of dollars and an infinite amount of hype is being poured in it, even though the huge AI bubble could burst soon.

Dr Emmanuel Maggiori's book, Smart Until It's Dumb: Why artificial intelligence keeps making epic mistakes (and why the AI bubble will burst) was published last year. In it, the author exposes unreasonable expectations, dubious practices and misinformation about artificial intelligence. The London-based software engineer has been working in the AI industry for 10 years and is on a mission to raise awareness of the deceptively simple mechanisms behind AI's remarkable achievements and expose the industry's hidden truths. At the event in Kolozsvár/Cluj--Napoca, Ákos Péter, a computer science student and member of the MCC University Program, welcomed the participants. The moderator of the discussion was Borbála Hatházy, Project Manager of MCC's International Relations team.

First of all, the speaker pointed out one of the major, and so far seemingly unfixable, flaws of ChatGPT: hallucination. This refers to cases where AI creates plausible inaccuracies or false information. But why does AI hallucinate in the first place? While humans have a comprehensive model of how the world works, current AI learns from examples: translation from documents written by humans, collects examples of how to autocomplete a sentence from the internet, and learning how to drive by processing images from inside a car paired with a human driver’s decisions. AI can learn shortcuts to imitate the job instead of learning the „world model”, so it cannot cope well with unexpected situations and unusual things.

Unfortunately, we are not getting rid of hallucinations any time soon. They are not bugs that can be fixed, but an expected feature of current machine learning methodology, and they cause problems in real, running projects. They are particularly visible in the case of self-driving cars, they’re killing the industry, but even a lawyer used ChatGPT in court and cited fake cases. The expert says it is futile to expect the hallucination issue to resolve itself quickly. A new AI method, closer to artificial general intelligence (AGI), needs to be found to eliminate the problem.

Emmanuel Maggiori advised that AI can be very useful in facilitating our work, but it is important to assess the needs: how complex is the task, what does the client want, are hallucinations okay with our task? For example, translating hotel reviews is ok, it is not expected for them to be perfect; but for legal or medical documents, it is essential. SEO-driven blog articles can be written by AI, but an excellent, exciting and valuable text requires the journalist's experience, research, the atmosphere and uniqueness they convey. Sometimes, AI’s work is good but not excellent. And sometimes acceptable quality is fine - these are the jobs that AI can and will replace. “Some jobs will be impacted, but the transition will be smoother than some people say. In order to future-proof your job, pick a career path where there’s a gap between good and excellent, and where excellent is appreciated”, the speaker added.

It is important to have a realistic picture of the capabilities and limits of artificial intelligence. Opinions are very divided: some compare it to the industrial revolution, others only see its dark side, but the picture is more nuanced. It will not eliminate all jobs, but it will make some irrelevant. Nor should it be feared, but the dangers need to be understood, and over-reliance and huge investments need to be better thought through. There’s a lot of craziness in the tech field, and a lot of enthusiasm, but there comes a day you have to pay the bills. So will the bubble burst? It may take a year or two for companies to see that it is not generating money as expected.