From: mk_thisisit
It is becoming increasingly difficult to identify tests where humans demonstrably outperform machines [00:00:00]. In many domains, artificial intelligence (AI) has already surpassed or shows strong potential to surpass human capabilities [00:07:23].
Current Superiority of AI
Medical Field
Studies involving AI chatbots like GPT-4 on medical topics have shown that the AI significantly outperformed groups of doctors in both the quality of advice given and empathy towards patients [00:08:32]. Most doctors were found to be very weak or lacking in empathy [00:09:00]. The expectation is that these systems could relieve human professionals, allowing them to focus on areas currently overburdened [00:09:06].
Programming
While complex systems requiring human-client communication may still favor humans, in programming itself, AI is proving to be faster, better, and capable of correcting human mistakes [00:07:38]. Programmers now frequently consult AI systems to verify their code [00:08:05].
Image Recognition and Emotional Interpretation
AI systems now demonstrate superior image recognition [00:08:21]. Furthermore, AI is better at interpreting emotional states based on video recordings, even teaching humans about their own emotions, which is described as “completely surprising” [00:12:08].
Understanding Animal Communication
Professor Włodzisław Duch states that strong artificial intelligence will eventually help humanity understand animals [00:00:15]. Efforts are underway to analyze sounds made by whales in their songs or dolphins’ squeaks to detect symbols and understand their communication’s meaning by relating them to their behavior [00:13:10]. The goal is to eventually communicate with animals in their own language [00:13:39]. This capability of AI to “rebuild bridges” extends even to animal relationships [00:13:52].
Evolving Capabilities of AI
AI Creativity and “Temperature”
AI systems regulate their creativity through a parameter known as “temperature” [00:00:45], [00:04:10]. This parameter determines how much the AI can deviate from learned patterns, allowing it to generate new, varied, and even “confabulated” responses [00:04:16], [00:05:11]. High temperature allows for more distant associations and fantasizing [00:06:36]. This concept is analogous to neuronal noise in the human brain, which causes human responses to vary [00:04:41].
Stanisław Lem’s Predictions and AI Reality
Stanisław Lem, a Polish writer, was notably skeptical about AI progress in his time, particularly regarding machines developing consciousness or genuine thought [00:01:30], [00:01:45]. He distinguished between “intelligence” (e.g., chess-playing) and “reason,” linking reason to consciousness [00:02:01]. However, two decades later, AI capabilities have surprised everyone, including Lem himself [00:01:40], [00:02:56]. The belief that machines couldn’t think or possess a “spark of intelligence” has not come true [00:01:51], [00:01:56].
Lem also speculated on the idea that information could be the basis of matter and the universe, a concept now considered by physicists [00:03:07]. While some of his predictions proved highly visionary, he himself was critical of his ability to be a visionary [00:02:50].
AI as a Tool User
Unlike the human brain, which has specialized structures for different functions, AI systems often have a single, massive network [00:10:38]. However, AI can use various tools, potentially more widely than humans, by integrating different sensors (e.g., infrared, ultrasound) [00:11:04].
AI in Robotics
Recent advancements show huge progress in robotics through the integration of robotics and artificial intelligence [00:18:01]. Robots are now equipped with speech analysis and discussion capabilities [00:18:23].
One notable example is the Boston Dynamics “Spot” robot dog. When asked about its parents, it led to a room with old robots and stated, “This is his line of ancestors” [00:18:53]. This surprised its creators as nothing like it was programmed [00:19:03]. This indicates that current AI systems are not merely repeating learned information but are creative, generating new statements based on their perceptions and internal neural network states [00:19:17]. They operate as open systems, not subject to the limitations of classical symbolic systems [00:19:28].
Abstraction and Internal States in Robots
AI systems are moving to higher levels of abstraction [00:17:00]. For instance, a robot shown many animals it has learned to recognize, when asked to “look for a dinosaur,” can infer it’s an animal and choose a unique, previously unknown one, identifying it as a dinosaur [00:16:35].
Moreover, extensive cooperation among institutions has led to collecting information about the internal states of robots. This allows for improving robot operation in the world, including real-time systems (RTX) that enable robots to understand their own body, distance, and regulate forces, akin to human internal sensations [00:17:04], [00:17:22]. This internalizing of sensations will drive significant progress in robotics [00:17:54].
Addressing Challenges and Future Prospects
AI is also expected to help filter the overwhelming amount of information on the internet, which Lem noted was a problem [00:14:00], [00:14:08]. It can also assist with education, serving as an “ethical system” to protect humans from their own “stupidity,” as Lem once described [00:14:55], [00:15:01].
The future of AI is not necessarily terrible; it is likely that AI can be harnessed to create a more human-friendly environment [00:15:23]. The most surprising recent development in AI development has been the release of GPT chat, which allowed for natural conversation and rapid expansion into hundreds of languages [00:15:52].