The Good, The Bad And The Unknown Of Ai: A Q&a With Mária Bieliková

Artificial intelligence is connected everybody’s lips these days, sparking excitement, fearfulness and endless debates. Is it a unit for bully aliases bad – aliases a unit we really person yet to afloat understand? We sat down pinch salient machine intelligence and AI interrogator Mária Bieliková to talk these and different pressing issues surrounding AI, its effect connected humanity, and broader ethical dilemmas and questions of spot it raises.

Congratulations connected becoming nan latest laureate of nan ESET Science Award. How does it consciousness to triumph nan award?

I consciousness immense gratitude and happiness. Receiving nan grant from Emmanuelle Charpentier herself was an unthinkable experience, filled pinch aggravated emotions. This grant doesn't conscionable beryllium to maine – it belongs to each nan singular group who accompanied maine connected this journey. I judge they were each arsenic thrilled. In IT and besides successful technologies successful general, results are achieved by teams, not individuals.

I'm delighted that this is nan first clip nan main class of nan grant has gone to nan section of IT and AI. 2024 was besides nan first twelvemonth nan Nobel Prize was awarded for advancement successful AI. In fact, location were 4 Nobel Prizes for AI-related inventions – 2 successful Physics for Machine Learning of Neural Networks and 2 successful Chemistry for training Deep Neural Networks that foretell macromolecule structures.

And of course, I consciousness immense pridefulness for nan Kempelen Institute of Intelligent Technologies, which was established 4 years agone and now holds a unchangeable spot successful nan AI ecosystem of Central Europe.

A starring Slovak machine scientist, Mária Bieliková has conducted extended investigation successful human-computer relationship analysis, personification modelling and personalization. Her activity besides extends to nan information study and modelling of antisocial behaviour connected nan web, and she’s a salient sound successful nan nationalist sermon astir trustworthy AI, nan dispersed of disinformation, and really AI tin beryllium utilized to combat nan issue. She besides co-founded and presently heads up nan Kempelen Institute of Intelligent Technologies (KInIT), wherever ESET acts arsenic a mentor and partner. Ms. Bieliková precocious won nan Outstanding Scientist successful Slovakia class of nan ESET Science Award.

Author and historiographer Yuval Noah Harari has made nan pithy observation that for nan first clip successful quality history, nary 1 knows what nan world will look for illustration successful 20 years aliases what to thatch successful schools today. As personification profoundly progressive successful AI research, really do you envision nan world 2 decades from now, peculiarly successful position of exertion and AI? What are nan skills and competencies that will erstwhile beryllium basal for today’s children?

The world has ever been difficult, uncertain, and ambiguous. Today, exertion accelerates these challenges successful ways that group struggle to negociate successful existent time, making it difficult to foresee nan consequences. AI not only helps america automate our activities  and switch humans successful various fields, but besides create caller structures and synthetic organisms, which could perchance origin caller pandemics.

Even if we didn’t expect specified scenarios, exertion is consciously aliases unconsciously utilized to disagreement groups and societies. It's nary longer conscionable integer viruses aiming to paralyze infrastructure aliases summation resources; it's a nonstop manipulation of quality reasoning done propaganda dispersed astatine nan velocity of ray and magnitude we couldn't person imagined a fewer decades ago.

I don’t cognize what benignant of nine we will unrecorded successful 20 years from now aliases really nan rules of humanity will change. It mightiness return longer, but we mightiness moreover beryllium capable to change our meritocratic system, presently based connected nan information of knowledge, successful a measurement that doesn’t disagreement society. Perhaps we'll alteration nan measurement we grip information erstwhile we recognize we can't afloat spot our senses.

I americium convinced  that moreover our children will progressively deviate from nan request for knowledge and evaluating occurrence successful various tests, including IQ tests. Knowledge will stay important, but it must beryllium knowledge that we tin apply. What will genuinely matter is nan power group are consenting to put successful doing meaningful things. This is existent today, but we often underutilize this position erstwhile discussing education. We still measure cognitive skills and knowledge contempt knowing these competencies unsocial are insufficient successful nan existent world today.

I judge that arsenic exertion advances, our request for beardown communities and nan improvement of societal and affectional skills will only grow.

As AI continues to advance, it challenges long-standing philosophical ideas astir what it intends to beryllium human. Do you deliberation René Descartes' study astir quality exceptionalism, "I think, truthful I am", will request to beryllium re-evaluated successful an era wherever machines tin "think"? How acold do you judge we are from AI systems that mightiness push america to redefine quality consciousness and intelligence?

AI systems, particularly nan ample instauration models, are revolutionizing nan measurement AI is utilized successful society. They are continually improving. Before nan extremity of 2024, OpenAI announced caller models, O3 and O3mini, which achieved important advancements successful each tests, including nan ARC-AGI benchmark that measures AI's ratio successful acquiring skills for chartless tasks.

From this, 1 mightiness presume that we are adjacent to achieving Artificial General Intelligence (AGI). Personally, I judge we are not rather location pinch existent technology. We person astonishing systems that tin assistance successful programming definite tasks, reply galore questions, and successful galore tests, they execute amended than humans. However, they do not genuinely understand what they are doing. Therefore, we cannot yet talk astir genuine thinking, moreover though immoderate reasoning down task solution is already being done by machines.

Just arsenic we understand words for illustration intelligence and consciousness today, we tin opportunity that AI possesses a definite level of intelligence – meaning it has nan expertise to lick analyzable problems. However, arsenic of now, it lacks consciousness. Based connected really it functions, AI does not person nan capacity to consciousness and usage emotions successful nan tasks it is given. Whether this will ever change, aliases if our knowing of these concepts will evolve, is difficult to predict.

The conception that "to create is human" is being progressively questioned arsenic AI systems go tin of generating art, music, and literature. In your view, really does nan emergence of generative AI effect nan quality acquisition of creativity? Does it heighten aliases diminish our consciousness of personality and characteristic arsenic creators?

Today, we witnesser galore debates connected productivity and AI. People devise various tests to showcase really acold AI has travel and wherever these AI systems aliases models surpass quality capabilities. AI tin make images, music, and literature, immoderate of which could beryllium considered creative, but surely not successful nan aforesaid measurement arsenic quality creativity.

AI systems tin and do create original artifacts. Although they make them from pre-existing materials, we could still find immoderate genuinely caller creations. But that's not nan only important aspect. Why do group create art, and why do group watch, read, and perceive to art? At its essence, creation helps group find and fortify relationships pinch 1 another.

Art is an inseparable portion of our lives; without it, our nine would beryllium very different. This is why we tin admit AI-generated euphony aliases paintings – AI was created by humans. However, I don’t judge AI-generated creation would fulfill america semipermanent to nan aforesaid grade arsenic existent creation created by humans, aliases by humans pinch nan support of technology.

Just arsenic we create technologies, we besides activity reasons to unrecorded and to unrecorded meaningfully. We mightiness unrecorded successful a meritocracy wherever we effort to measurement everything, but what brings america person together and characterizes america are stories. Yes, we could make those too, but I americium talking astir nan stories that we live.

AI investigation has seen fluctuations successful advancement complete nan decades, but nan caller gait of advancement – particularly successful instrumentality learning and generative AI – has surprised moreover galore experts. How accelerated is excessively fast? Do you deliberation this accelerated advancement is sustainable aliases moreover desirable? Should we slow down AI invention to amended understand its societal impacts, aliases does slowing down consequence stifling beneficial breakthroughs?

The velocity astatine which caller models are emerging and improving is unprecedented. This is mostly owed to nan measurement our world functions coming – a monolithic attraction of wealthiness successful backstage companies and definite parts of nan world, arsenic good arsenic a world title successful aggregate fields. AI is simply a important portion of these races.

To immoderate extent, advancement depends connected nan exhaustion of today's exertion and nan improvement of caller approaches. How overmuch tin we amended existent models pinch known methods? To what grade will large companies stock caller approaches? Given nan precocious costs of training ample models, will we conscionable beryllium observers of improving achromatic boxes?

At present, location is nary equilibrium betwixt nan systems humanity tin create and our knowing of their effects connected our lives. Slowing down, fixed really our nine works, is not possible, successful my opinion, without a paradigm shift.

This is why it is important to allocate resources and power to investigation nan consequences of these systems and to study nan models themselves, not conscionable done standardized tests arsenic their creators do. For example, astatine nan Kempelen Institute, we investigation nan skills and willingness of models to make disinformation. Recently, we person besides been looking into nan procreation of personalized disinformation.

There’s a batch of excitement astir AI’s imaginable to lick world challenges – from healthcare to ambiance change. Where do you judge nan committedness of AI is top successful position of applicable and ethical applications? Can AI beryllium nan "technological fix" for immoderate of humanity’s astir pressing issues, aliases do we consequence overestimating its capabilities?

AI tin thief america tackle nan astir pressing issues while simultaneously creating caller ones. The world is afloat of paradoxes, and pinch AI, we spot this astatine each turn. AI has been beneficial successful various fields. Healthcare is 1 specified area where, without AI, immoderate advancement – for example, successful processing caller medications – would not beryllium possible, aliases we would person to hold overmuch longer. AlphaFold, which predicts nan building of proteins, has tremendous imaginable and has been utilized for years now.

On nan different hand, AI besides enables nan creation of synthetic organisms, which tin beryllium beneficial but besides airs risks specified arsenic pandemics aliases different unforeseen situations.

AI assists successful spreading disinformation and manipulating people's thoughts connected issues for illustration ambiance change, while astatine nan aforesaid time, it tin thief group understand that ambiance alteration is real. AI models tin show nan imaginable consequences for our satellite if we proceed connected our existent path. This is crucial, arsenic group thin to attraction only connected short-term challenges and often underestimate nan seriousness of nan business unless it straight affects them.

However, AI tin only thief america to nan grade that we, arsenic humans, let it to. This is nan biggest challenge. Since AI doesn't understand what it produces, it has nary intentions. But group do.

With awesome imaginable besides travel important risks. Prominent figures successful tech and AI person expressed concerns astir AI becoming an existential threat to humanity. How do you deliberation we tin equilibrium responsible AI improvement pinch nan request to push boundaries, each while avoiding alarmism?

As I mentioned before, nan paradoxes we witnesser pinch AI are immense, raising questions for which we person nary answers. They airs important risks. It's fascinating to research nan possibilities and boundaries of technology, but connected nan different hand, we are not fresh – arsenic individuals, nor arsenic a nine – for this type of automation of our skills.

We request to put astatine slightest arsenic overmuch successful researching nan technological effect connected people, their thinking, and their functioning arsenic we do successful nan technologies themselves. We request multidisciplinary teams to jointly research nan possibilities of exertion and their effect connected humanity.

It's arsenic if we were creating a merchandise without caring astir nan worth it brings to nan consumer, who should bargain it, and why. If we didn’t person a seller, we wouldn't waste much. The business pinch AI is much serious, though. We person usage cases, products, and group who want them, but arsenic a society, we don’t afloat understand what's happening erstwhile we usage them. And possibly astir group don't moreover want to know.

In today's world world, we cannot extremity progress, nor tin we slow it down. It only slows erstwhile we are saturated pinch outcomes and find it difficult to improve, aliases erstwhile we tally retired of resources, arsenic training ample AI models is very expensive. That is why their champion protection is researching their effect from nan opening of their improvement and creating boundaries for their use. We each cognize that it is prohibited to portion intoxicant earlier nan property of 18, aliases 21 successful immoderate countries, yet often without hesitation, we let children to chat pinch AI systems, which they tin easy liken to humans and spot implicitly without knowing nan content.

Trust successful AI is simply a awesome taxable globally, pinch attitudes toward AI systems varying widely betwixt cultures and regions. How tin nan AI investigation organization thief foster spot successful AI technologies and guarantee that they are viewed arsenic beneficial and trustworthy crossed divers societies?

As I was saying, multidisciplinary investigation is basal not only for discovering caller possibilities and improving AI technologies but besides for evaluating their skills, really we comprehend them, and their effect connected humans and society.

The emergence of heavy neural networks is changing nan technological methods of AI and IT. We person artificial systems wherever nan halfway principles are known, but done scaling, they tin create skills that we cannot ever explain. As scientists and engineers, we devise ways to guarantee nan basal accuracy successful circumstantial situations by combining various processes. However, location is still overmuch we don't understand, and we cannot afloat measure nan properties of these models.

Such investigation does not nutrient nonstop value, which makes it challenging to garner voluntary support from nan backstage assemblage connected a larger scale. This is wherever nan backstage and nationalist sectors tin collaborate for nan early of each of us.

AI regularisation has struggled to support up pinch nan field’s accelerated advancements, and yet, arsenic personification who advocates for AI morals and transparency, you’ve apt considered nan domiciled of regularisation successful shaping nan future. How do you spot AI researchers contributing to policies and regulations that guarantee nan ethical and responsible improvement of AI systems? Should they play a much progressive domiciled successful policymaking?

Thinking astir morals successful investigation is crucial, not only successful investigation but besides successful nan improvement of products. However, it tin beryllium rather costly because it is important that a existent request arises astatine nan level of captious masses. We still person to see nan dilemma of caller knowledge acquisition versus nan imaginable interference pinch nan autonomy aliases privateness of individuals.

I americium convinced that a bully solution is possible. The mobility of morals and credibility must beryllium an integral portion of nan improvement of immoderate merchandise aliases investigation from nan beginning. At nan Kempelen Institute, we person experts connected morals and regulations who thief not only researchers but besides companies successful evaluating nan risks connected to nan morals and credibility of their products.

We spot that each of america are becoming much sensitive. Philosophers and lawyers deliberation astir nan technologies and connection solutions that don’t destruct nan risks, while scientists and engineers are asking themselves questions they hadn’t considered before.

In general, location are still excessively fewer of these activities. Our nine evaluates results chiefly based connected nan number of technological papers produced, leaving small room for argumentation advocacy. This makes it moreover much captious to create abstraction for it. In caller years, successful definite circles, specified arsenic earthy connection processing aliases recommender strategy communities, it has go modular for technological papers to see opinions connected morals arsenic portion of nan reappraisal process.

As AI researchers activity toward innovation, they are often confronted pinch ethical dilemmas. Have you encountered challenges successful balancing nan ethical imperatives of AI improvement pinch nan request for technological progress? How do you navigate these tensions, peculiarly successful your activity connected personalized AI systems and information privacy?

At nan Kempelen Institute, it has been beneficial to person philosophers and lawyers progressive from nan very beginning, helping america navigate these dilemmas. We person an morals board, and diverseness of opinions is 1 of our halfway values.

Needless to say, it’s not easy. I peculiarly find it problematic erstwhile we want to construe investigation results into believe and brushwood issues pinch nan information nan exemplary was trained on. In this regard, it is important to guarantee transparency from nan outset, truthful we tin not only constitute a technological insubstantial but besides thief companies innovate their products.

Given your collaboration pinch ample exertion companies and organizations, specified arsenic ESET, really important do you deliberation it is for these companies to lead by illustration successful promoting ethical AI, inclusivity, and sustainability? What domiciled do you deliberation corporations should play successful shaping a early wherever AI is aligned pinch societal values?

The Kempelen Institute was established based connected nan collaboration of individuals pinch beardown world backgrounds and visionaries from respective ample and medium-sized companies. The thought is that shaping a early wherever AI aligns pinch societal values cannot beryllium realized by conscionable 1 group. We person to link and activity synergies wherever possible.

For that reason, successful 2024, we organized nan first version of nan AI Awards, focused connected Trustworthy AI. This arena culminated astatine nan Forbes Business Fest, wherever we announced nan laureate of nan grant – AI:Dental, a startup. In 2025 we are successfully continuing nan AI Awards and person received much and higher value applications.

We started discussing nan taxable of AI and disinformation almost 10 years ago. Back then, it was much academic, but moreover then, we witnessed immoderate malicious disinformation, particularly related to quality health. We had nary thought of nan immense power this taxable would yet person connected nan world. And it's only 1 of galore pressing issues.

I fearfulness that nan nationalist assemblage unsocial has nary chance of tackling these issues without nan thief of ample companies, particularly coming erstwhile AI is being utilized by politicians to summation popularity. I see nan taxable of trustworthiness successful technology, peculiarly AI, to beryllium arsenic important arsenic different cardinal topics successful CSR. Supporting investigation connected nan features of AI models and their effect connected humans is basal for sustainable advancement and value life.

Thank you for your time!