From: mk_thisisit
Professor Marta Kwiatkowska is a distinguished Polish computer scientist renowned for her pioneering work and achievements in the field, particularly at the University of Oxford and in Great Britain [00:00:00].
Pioneering Achievements at Oxford
Marta Kwiatkowska was the first woman in the history of the University of Oxford to receive the title of Professor of Computer Science [00:00:07], [00:01:33]. She is also the first woman in the history of Great Britain to receive the Milner Prize, one of the main awards of the Royal British Scientific Society [00:00:11]. Years later, she also received a medal of the Royal Scientific Society in London, again as the first woman to be awarded it [00:39:00]. These achievements highlight her significant contributions and leadership in computer science and the broader recognition of women in science [00:00:07], [00:01:33], [00:39:00].
Views on Artificial Intelligence
Professor Kwiatkowska defines artificial intelligence (AI) today in the context of its historical origins, noting that the term was introduced over 50 years ago with the idea of creating a copy of human intelligence [00:02:04]. However, she asserts that current AI is artificial but lacks true “intelligence” in the human sense, as its definition has always been unclear [00:02:23], [00:02:47].
Limitations of Current AI
She emphasizes that computers are not yet capable of achieving human intelligence, a problem that is “very difficult” and currently not fully understood [00:00:27], [00:32:27]. She critiques the media and some scientific texts for misusing the term “intelligence” in relation to AI, especially when discussing “conscious artificial intelligence” or “strong artificial intelligence” [00:03:02], [00:03:41]. According to Kwiatkowska, what is currently seen, such as in large language models, is still based on statistics and interpolation, which is an “illusion” of understanding [00:52:03], [00:01:00], [00:10:09], [00:10:32].
She points out several key limitations:
- Lack of Human-like Understanding and Learning: Current models operate on “pattern matching” rather than genuine understanding, which is a core element of human intelligence that remains unknown how it works [00:13:04], [00:13:33]. Human learning, especially in children, is fundamentally different from how language models are trained [00:13:50], [00:23:47]. Humans also rely on “mental models” when processing information, which language models currently lack [00:24:10].
- Sensory Input: AI systems, like autonomous taxis, currently lack the full range of human senses (e.g., smell, gestures, spatial understanding) which are crucial for navigating the world intuitively and making complex decisions [00:19:15], [00:19:52]. They may require additional sensors to achieve real copies of human intelligence [00:20:51].
- Hallucinations: A significant issue in large language models is “hallucinations,” where models generate incorrect or fabricated information. This stems from their artificial, infinite search space and interpolation methods [00:12:20], [00:26:10]. To combat this, a two-level system is proposed where LLMs handle communication, and a backend handles factual accuracy [00:27:05].
- Data vs. Intelligence: Kwiatkowska believes that progress in AI is not about having “more and more data” but rather about “intelligently using this data” and integrating physical models of the world [00:25:05], [00:29:50].
- Autonomous Taxis: She cites examples of autonomous taxis in San Francisco stopping when unsure what to do, causing protests, and one even driving into Chinese New Year celebrations, demonstrating their lack of human-like contextual understanding and intuition [00:01:00], [00:07:50], [00:08:41]. This highlights the need for AI to behave more like a human driver [00:09:15].
Future Development and Expectations
Professor Kwiatkowska expects greater autonomy from AI systems in the future, leading to “better, more safer [sic], accurate” decisions [00:07:03]. However, achieving human-like intelligence, including understanding context and intuition, will take “much longer than 5-10 years” [00:00:31], [00:09:45]. She suggests a need for a “different approach” to teaching AI systems, perhaps involving more collaboration between computer scientists, psychologists, and neuroscientists [00:00:35], [00:21:46], [00:30:09]. She believes that AI does not need to be an exact copy of human intelligence, but rather “enough such intelligence that corresponds to the application” [00:14:27].
Research and Innovations
Professor Kwiatkowska and her group have developed a tool called Pryzm (also referred to as Prism), which is used for modeling systems [00:00:45], [00:37:44]. This system is particularly designed for stochastic models, where probabilistic actions are selected [00:38:29]. She has worked on this tool with her students for over 20 years [00:38:42]. For their work on Pryzm, they received the PRISM Test of Tool Award [00:38:47].
Her current research focuses on multi-agent artificial intelligence systems. These models allow agents to perform actions and communicate using perception mechanisms, such as neural networks for automatic cars [00:34:31], [00:35:05]. The goal is to program automatic vehicles to understand complex situations, like predicting pedestrian behavior, similar to how human drivers make predictions [00:35:43], [00:36:01].
Career Path and Recognition
Professor Kwiatkowska began her academic journey at the University of Warsaw, being among the first years of computer science students [00:36:58]. She started as an assistant there before moving to England to pursue her doctorate [00:37:07]. In England, she secured a position as a lecturer, equivalent to an assistant professor, and later became a full professor [00:37:12]. A breakthrough moment in her scientific career was when she started her own research group and began developing the Pryzm tool [00:37:40].
Reflections on Polish Science and Commercialization
Professor Kwiatkowska observes significant progress in Polish science, noting an increase in research groups and publications in top conferences and journals [00:39:32]. However, she points out that the scientific career path in Poland has not changed much in over 40 years, with positions not being openly advertised, which is not standard in the West [00:40:01]. Additionally, assessment of achievements in Poland still relies on point systems, unlike the West where successes in top conferences and journals are prioritized [00:40:28]. These observations offer insights into the impact of cultural and educational factors on Polish programming talent and the education and global impact of Polish programmers.
Regarding the commercialization of scientific projects, particularly at universities like Oxford, there are no direct obligations for scientists to commercialize their work [00:41:23]. However, the government encourages scientists to consider applications beyond basic research [00:41:27]. While some, like Kwiatkowska, prefer to publish their tools as Open Source, others may choose to develop closed-source commercial software [00:41:44]. The University of Oxford has systems and regulations in place to support commercialization, but it requires substantial additional effort from scientists who must balance scientific work with business endeavors [00:42:19].