Today, many of the world's tech companies are in a unique race to literally breathe life into artificial intelligence (AI). Machine learning systems have already become an integral part of business for many, so it is not surprising if the news about AI and neural networks comes across your eyes almost every day. Usually the headlines of such news sound like this: "AI defeated a person in a video game" or "AI imitates human speech", and sometimes "AI is more effective than a person in determining the development of cancer." Are we really close to the moment when the intelligence of machines can be compared with humans, or to the moment when man and machine can conduct small talk and work together as naturally as humans do among themselves? Are machines far from self-awareness?
Premature hype
While all of the above AI headlines are already real, people like Yang LeCun, Director of AI Development at Facebook and Professor of Computer Science at NYU, believe that we are overestimating the capabilities of today's AI systems and creating an overly hype around them.
“In fact, we are still far from creating machines that can learn basic worldview in the same way that humans and animals can do it,” LeCune commented in an interview with The Verge.
"Yes, you can't argue that in some isolated areas machines have already acquired capabilities that surpass human ones, but in the prospect of creating a general universal artificial intelligence, we have not even come close to the level of a rat."
The so-called general artificial intelligence is a system that does not require the participation of a human operator and is capable of performing almost any task that a human can perform. The current AI systems are specialized and can work only with one or another task, for example, to deal with speech or image recognition or highlight specific objects in huge amounts of data, that is, do only what they were programmed to do. Such specialized AIs are often referred to as "applied AIs" or "highly specialized AIs", which only once again shows their limited intellectual capabilities.
LeCun's claims are agreed by Manuel Sebrian, an MIT employee and co-developer of Shelley, an AI algorithm capable of writing scary stories.
Promotional video:
“AI is just a great tool. But in my opinion, based on my experience with Shelley, AI is very far from being able to create horror stories on a professional level, as it is still very far from the level of human intelligence,”notes Sebrian.
LeCune generally believes that, despite all the amazing level of progress that researchers and developers of artificial intelligence have achieved in recent years, working with machine learning and neural networks is not exactly the development of the very real artificial intelligence that everyone is dreaming of today. …
“I in no way want to belittle the merits of our fellow engineers and researchers from DeepMind who work with the same AlphaGo, however, when people interpret the improvement of AlphaGo as a significant advance in the development of general artificial intelligence, it is wrong. Because it’s not at all like that,”says LeCun.
Pierre Barot, CEO of Aiva Technologies, which developed Aiva's AI algorithm for music creation, also believes that the progress we have made in creating synthetic intelligence is somewhat exaggerated.
“General artificial intelligence is a topic that is getting a lot of attention. In general, I am, of course, optimistic about how quickly technology is developing, but at the same time I think that most people simply do not understand the complexity of our own brain, not to mention how difficult it will be to create an artificial one,”says Baro …
Building general artificial intelligence
Today people are very fond of using the term AI for any reason, even where the conversation may be about something completely different. In any news about AI, you can come across terms like "machine learning" or "deep learning", as well as "neural networks". Although each of these terms is to some extent related to AI, in fact, we are not talking about AI as such.
Machine learning is a tool. A set of algorithms that make up an intelligent system that learns by absorbing a huge amount of data. Also, deep learning is a type of machine learning, not necessarily tied to a specific task. On the other hand, a neural network is a system that mimics the work of the brain, but again it works only within the framework of the features according to which machine learning algorithms are created.
AI experts believe that all three of the above components are the fundamental basis for creating a synthetic intelligence with the ability to think in a human way, that is, to be aware of one's actions and their consequences. But we are only at the very beginning of this path. No, we have indeed made a lot of progress, but current developments have hardly moved us from place towards creating real intelligence. Nevertheless, it is quite interesting to wonder when we should expect this type of artificial intelligence to emerge. Is there a time frame?
According to Luke Tang, head of AI startup TechCode, the real shift towards full-fledged artificial intelligence "will begin with a breakthrough in the development of unsupervised machine learning algorithms." Once we get there, "machine intelligence will very quickly surpass human intelligence," Tang shared in an interview with Futurism.
To say that this will be difficult to achieve is practically nothing to say.
“To create a full-fledged general artificial intelligence requires serious progress not only in software development. Significant advances in neuroscience and hardware development are needed,”says Baro.
“We are at the edge of Moore's Law when transistors get so small that they simply cannot be physically obtained. And new hardware environments, such as quantum computing, have not yet been able to demonstrate their superiority over our conventional hardware, even when performing standard tasks,”the expert adds.
Many agree that in order to be able to consider AI as true intelligence, it must cope with solving five specific problems, the first of which is the Turing test, where a machine needs to convince a person that he is talking to another person, not a machine. … The same Baro is convinced that the current generation will be able to witness how AI will successfully cope with the Turing test, that is, in fact, deceive a person. Nevertheless, the expert believes that "it will not necessarily be just general artificial intelligence, but something that is already closer to it."
Intelligence enhancement
It is impossible not to note that the emergence of general artificial intelligence will be the harbinger of the so-called technological singularity. For those who have forgotten, recall that the concept of technological singularity speaks of the moment when intelligent machines surpass the human level of intelligence, stimulating a rampant and exponential technological growth that, at a fundamental level, promises to transform our lives. The author of the term, like the entire concept, is Vernor Vinge, who wrote the following in 1993:
“We will soon be able to create an intelligence that surpasses our own. When this happens, human history will reach a kind of singularity, an intellectual transition to a new level. It will be impossible to escape from it, just as it is impossible to escape from the center of a black hole. From that moment on, the world will begin to change so much that it goes beyond our understanding."
Despite the fact that this moment is something that people like SoftBank CEO Masayoshi Son and futurist Ray Kurzweil are eagerly awaiting, there are those (like Elon Max, Stephen Hawking and even Bill Gates) who clearly do not I'm very happy about this prospect. They argue that people simply do not understand what it really means to get artificial superintelligence, and we are clearly not ready for the possible consequences that a technological singularity can bring.
But why is it necessary to look at the question from this point of view? Why is it imperative to consider artificial intelligence as a sunset for all mankind, and not as its companion, friend, assistant, in the end? Musk, in fairness, is considering such an idea, which is why he created the Neuralink project. Kurzweil mentioned this cooperation between man and machine when he said that in the future nanobots will live inside us, which will greatly improve our capabilities.
“We need to focus on the benefit that AI can give us - enhanced intelligence, that is, human intelligence, whose capabilities will be enhanced by AI,” says Baro.
Algorithms like Aiva and Shelley are already showing their benefits in working with people. At the same time, intelligent robots like Hanson Robotics' Sophia and SotfBank's Pepper make it easy for us to imagine that truly intelligent machines are already among us. Perhaps superintelligence with an IQ of 10,000 as imagined by Masayoshi Son will actually become the very cognitive machine intelligence we all aspire to? If this is the case, then we will not have much to wait - about three decades, experts say.
“We will reach this level of AI, perhaps in 30-50 years. On the one hand, it may seem that this is a very long time, but on the other, it means that many of us will have a chance to live up to this moment, Tang concluded.
Nikolay Khizhnyak