is a data scientist with PhD in Physics, currently working in IoT
branch. He is passionate for artificial intelligence with ten years of
experience in automated data analysis and machine learning.
JUnQ: The everlasting technological progress is aimed to fulfill many needs of humans: most of them are physical, informational and commercial. In particular, robots were created to perform tasks that were too dangerous for humans or that humans could not or did not want to do. But what do we need intelligent machines for and what is implied by “Artificial Intelligence” (AI)?
Anton Bogomolov: The answer was already said – we need AI to make our life simpler, i.e. to simplify some routine work that humans have to do. Generally, we are heading towards automation, and in the ideal case, we want to automize everything, every kind of work. So far, the processes we are capable of automizing have been prioritized.
Now, what is understood by the term “AI”? Over the course of this interview we will go deeper in the discussion, so let’s start with a fairly broad definition: AI is something that is able to accomplish certain tasks with the help of self-learning.
JUnQ: Does it imply that AI is not meant to create anything, like art or music?
Anton Bogomolov: There is a number of definitions of AI. Indeed, the term “intelligence” implies that it can do creative work as well. It is not a simple calculator. You don’t just tell it what you want it to calculate, and then it does exactly what has been asked. It does something more complicated and, thus, it also involves some learning experience. In this context, the creative work does not necessarily mean being an artist or a musician, or a composer. A chatbot, as an example of an AI feature, is also a kind of creative work, because it is required to react accordingly or ask appropriate questions, in other words to be engaged into a conversation as a human would be i.e. express creativity.
Generally, yes, AI can generate art. For example, “Deep Dream”1 was popular a few years back. This algorithm uses AI to generate the dream-like appearance of the uploaded images. Another one is “Neural style transfer”2 which allows one to compose an image in the style of another image. Should one ever want to paint like Van Gogh or Picasso, this can be easily done, using this algorithm. There is also AI-composed music already creeping into the background of games, film, and media. With AI it is now possible to create music in different genres just at the push of a button.
JUnQ: In the news or podcasts, the term “machine learning” often seems to come together with AI. What is, simply put, machine learning and how does it relate to AI?
Anton Bogomolov: As I mentioned before there are many definitions of AI. In simple words, AI is a broader term than the machine learning (ML), i.e. AI includes ML. Being sort of an advanced algorithm, AI achieves specific goals by means of ML, at the same time it is able to adapt to its environment, just like humans. ML is also an algorithm, but a simpler one, with the key feature – the ability to learn (thus the name). It is not meant to achieve a global goal, its goal is to eventually enable programs to automatically improve through experience, without the programmer having to change the code. ML relies on working with data sets, that one needs to input first. It then examines and analyses the data to find common patterns, so that eventually it becomes possible to make experience-driven predictions or decisions.
JUnQ: So what it means is that AI does not exist without ML?
Anton Bogomolov: Right. Machine learning is a subset of AI, more like a tool to achieve AI. One example might be the first chatbots from the 90s. They had hardcoded “intelligence”, i.e hardcoded answers to possible questions. If such bot sees certain keywords it outputs accordingly relevant keywords. These did not have machine learning. But the intelligence of these was doubtful since the algorithm did not adapt. And as we discussed previously the key asset of AI is the ability to adapt.
JUnQ: Since we are on this page, how can one tell the difference between the AI system and a more “conventional” program?
Anton Bogomolov: There are “intelligence” tests for AI, among which the most renown one is the Turing Test.3 But this is more to test whether or not a system is capable of thinking like a human being. However, no AI technology today has passed the Turing test, i.e. that has shown to be convincingly intelligent and able to think. So, this is the main goal of this AI branch – we want to create a machine that will be indistinguishable from a human, in particular, that will be self-aware and act somewhat mindfully. In the end, such a machine will be able to pass the Turing test. Once again, so far, they do not exist. Self-awareness tuned out to be tough to realize.
Now, back to what was asked. I believe, no one is interested in differentiating AI from a mindless linear algorithm. Because as long as the desired goal is achieved no one cares what type of algorithm was used for it.
JUnQ: AI is no longer a futuristic concept, as some may naively think. Can you name some examples where is AI being used already? Are there any AI applications used in the everyday life of ordinary people?
Anton Bogomolov: The most straightforward example is our smartphones. The more recent ones can recognize the owner’s face. This is known to use neural networks. Also, in smartphones, there is Google assistant. Spoken inquiries are transferred to a server where neural network-based algorithms convert them to text, and which is then processed to deliver the relevant information. These are the simplest examples. We all watch Youtube where based on one’s watch history the system suggests what else one might be interested in. These AI-based recommendation engines now seem to know us to an uncanny degree.
If we now go further from everyday life, I would say AI is used pretty much in every field. In finance – there are already automatic trading robots. Some use AI for analysing financial markets to generate profitable trading strategies or make market predictions.
Autonomous driving has become very popular recently. There are even toys for children that make use of a variety of AI and ML technologies, including voice and image recognition, to identify the child and other people around, based on their voices and appearance. This all is owing to the computation power we currently have, which has advanced in the last years.
AI has found its application in medicine as well. As AI demonstrated remarkable progress in image-recognition tasks it is now widely used in medical radiology and computer tomography. One example is that there are neural networks that are trained to analyze tumours and do it as well as the top-class specialists in the field. Just as radiologists are trained to identify abnormalities based on changes in imaging intensities or the appearance of unusual patterns, AI can automatically find these features, and many others, based on its experience from the previous radiographic images, coupled with data on clinical outcomes. This also yields a more quantitative outcome, while radiologists perform only a quantitative assessment.4
JUnQ: As AI develops further is it going to make human jobs obsolete? And what will people be doing if there is nothing else to do?
Anton Bogomolov: Ideally, this is what we aim for – to have everything automized. But this can be achieved, in my opinion, only when so-called artificial general intelligence is realized. This will be a machine capable of experiencing consciousness and think autonomously and thus will be able to accomplish any intellectual task that a human being can.
What will happen to humans after all? There is a concept of universal basic income. The idea is that the robot replacing you is working on your behalf and you are given an income sufficient to meet basic needs, with zero conditions on that income. Because in the end the job is being done and the resources are being produced while you are free for other pursuits.
There has been a lot of research interest in this regard. Back in the 60’s, there was a researcher, John Bumpass Calhoun, who reported on an experiment with rats, the experiment is also known as “Universe 25”. The researchers provided rats with unlimited resources, such as water and food. Besides, they eliminated the danger otherwise coming from nature, like predators, climate, etc. Thus, the rats were said to be in “rat utopia”. At first, the population peaked but shortly after it started to exhibit a variety of abnormal, often destructive behaviours. After some time of the experiment, the rats became too lazy to reproduce and the population was on its way to extinction. There is, of course, the controversy over the implications of the experiment but it can be perceived as one of the possible scenarios of the future.
JUnQ: What about the programming jobs? And scientists?
Anton Bogomolov: Well, first we automize what we can do – so far, the simplest work. AI is now partly replacing the jobs of translators and customer service work. The next in line are self-driving cars that will automize the entire transportation industry, bus, and taxi drivers and so on. But programming jobs are of a different kind, they are creative. Programs that develop other programs exist already, but they are rather limited in what they can do.
Eventually, all jobs will be replaced. Programming jobs will be among the last ones though. Just as other creative jobs, including scientists.
One day we will have a super-intelligent machine, that develops further programs similar to itself at less expense and much faster compared to when supervised by humans. At some point we might not be able to follow its advances anymore and here comes the term “technological singularity”. This is believed to occur when AI starts discovering new science at enormous rates while always learning and evolving on top of it uncontrollably from human’s side.
JUnQ: Is the “singularity” inevitable?
Anton Bogomolov: There is an everlasting argument whether at all it is possible to realize a self-aware AI, that will act mindfully, much like a human. Therefore, depending on “yes” or “no” there will be a technological singularity or not. It can as well occur for other reasons, it is just that among others AI is more likely to bring us to the technological singularity.
On the other hand, it is not proven that such AI can ever be created, to be able to run autonomously and replace all of us. In this case, there will be no AI-induced singularity.
So, this is now a really hot topic in the community.
JUnQ: Does it mean that self-awareness is prerequisite for a possible singularity to occur and we are not yet passed the point of no return?
Anton Bogomolov: Right. The algorithms that exist now and are known to beat the world-class champions in chess and Go are harmless. They are just trained extraordinary well on one particular subject, to achieve a well-defined goal. They are not able to think outside of the box, like “what else is there that I could do”.
Once we create a machine that will be able to think this way, to exhibit human-level consciousness, it is expected to bring us to the singularity. Because it will be able to operate and develop without any supervision. All existing AI technologies do develop themselves but only to a certain degree, they do not have this freedom yet.
JUnQ: Speaking about self-awareness. For example, Sophia – the social humanoid robot developed by Hanson Robotics – realizes itself (herself) as being a programmed female robot. Does it mean that she is self-aware? How did they manage to program “her” self-realization?
Anton Bogomolov: As far as I understand she is programmed to answer this way. If there comes a question about what she thinks she is, her answer will be according to what has been built in her program. Most likely she was trained on thousands of real dialogs among people about their self-awareness. Like other AI systems, she also has machine learning that, if you feed it with enough data, will enable her to learn how to answer and how to behave, as people would.
Sophia communicates very well on a topic known in advance. Because in this case she can get trained in advance: they provide her with enough information about a given topic to get trained. Then she is able to have a sensible conversation because she has the statistics on what is typically answered when. Nevertheless, it is not as simple as when you say X, she replies Y. Thanks to machine learning what she says is a result of rather complicated non-linear connections.
I did not have a chance to speak with her personally though, but I think she is certainly not self-aware. Otherwise, the singularity would have been just around the corner by now. If she had a human-level consciousness, there would be nothing that she would need people for. She would be able to program herself to increase her memory. In just a few days she would reach the level of intelligence of all the people on Earth. In a few more days we would not be able to comprehend what level of intelligence she would have – again the exponential progress.
So, there is nothing we should worry about. She is still just a robot – more about illusion than intelligence. The shocking effect is also due to the fact that she looks like a human, has emotions and facial expressions. This unique combination of her features might make us a bit alert. And for sure Sophia is a great representation of all the advances of AI technology.
In fact, to able to realize human-level AI we essentially need to model a human’s brain. The human brain contains around 10 neurons. On the other hand, functional neural networks have in the order of tens of millions of neurons. These four orders of magnitude difference are sizeable. Moreover, it also takes quite some time to train a system with a large number of neurons. At the end of the day, we do not yet have the capacity to realize a human-level AI.
JUnQ: In case something goes wrong, will we able to “unplug” the machine. Do autonomous AI systems exist yet?
Autonomous systems do exist. Think of a toy-dog, that we have discussed already, or a vacuum cleaner, they are programmed to charge when needed. These are completely autonomous as long as the power source is available. Military branch sure has got some as well. I can imagine an armed flying drone, self-charging, and self-rechargeable.
But the existing autonomous AI systems are not a threat to humans. Despite having all the advantages of machine learning they follow a defined program to accomplish a specific task. It can be the best in recognizing people’s faces, shooting targets or avoiding bullets. But it is still a mindless machine, that we can destroy, or fool or at least hide all the power stations from it.
As long as any of these do not have human-level intelligence, as long as they are not smarter than us, they should not be considered as a potential threat.
JUnQ: So reaching human-level intelligence would be the point from which on AI can potentially live without us.
Anton Bogomolov: Correct. There is an opinion that biological life is just a means to create an electronic life. In other words, some believe that this is our mission, to give birth to an electronic conscious creature, surpassing our capacity, that will develop much faster than humans. In some sense, it is similar to the early times of our planet. Life on Earth began relatively early. But the first living creatures – unicellular organisms – were progressing very slowly, until the multicellular organism occurred, which boosted the progress tremendously. And the progress always seems to be exponential. Thus, the idea of this theory is that we create something to keep up to this exponential progress. And if we look at it globally, like in the scale of the Universe, if this should ever happen that AI takes over the world, it would make sense. Because AI would go further exploring the Universe much faster than we would. Thus, from the point of view of global progress, it would be more advantageous.
JUnQ: Now, when you put it this way the technological singularity does not sound so frustrating anymore. Are you optimistic overall? Will we make it to the end of the 21st century?
Anton Bogomolov: To me, it feels great to witness the progress and to be a part of it. But we will see how it goes. We live within a self-organized system, where everything has got a direction to go. Even though humans are all independent creatures, we still obey the same laws of synergy, we self-organize as well, we cluster forming cities, etc. And sure we also have something to move towards, thus we develop and evolve. So, this progress is so natural.
In fact, experts expect the technological singularity to occur already in the 21st century. But it is not trivial to give a correct estimate. On the other hand, not related to AI, there is research going on in the field of so-called negligible senescence. The idea is that by engineering the reversal of all the major molecular and cellular changes that occur with age we would enable us to constantly rejuvenate ourselves. The researchers believe that negligible aging for humans will be achieved in this century. There even exists a provocative opinion that the first human beings who will live to 1,000 years old are already alive.
At the end of the day, there has been tremendous progress in many fields, not only AI. Along with AI, we may succeed in developing other technologies, which will help us to prolong our lives as well as humans’ in general.
JUnQ: Thank you very much for the interview!
— Mariia Filianina
|||L.A. Gatys, A.S. Ecker and M. Bethge arXiv1508.06576 (2015).|
|||A. Hosny, C. Parmar, J. Quackenbush, L.H. Schwartz and H.J.W.L. Aerts Nature Reviews Cancer 18, 500 (2018).|