The future is at hand – smart AI chat bots kick off the new era. or not? Come learn how to prepare your business for global changes.Q
uote from a future novel? As a heading of Breaking News of May 2017. Google’s AI Alpha Go has won 3: 1 against the world’s most famous gamer after learning 1 year. Note that the machine had no pre-installed algorithms. It was to be learned from scratch.
All nodes at all levels of their deep neural network process millions of combinations of possible moves every second.
His memory stored thousands of games and helped him predict the consequences of each action he performed. He learned very hard. And he has won.
He could beat a human in any other activity. . for now.
Nowadays, neural networks allow machines to use reinforcement learning, neural backpropagation, and other complex algorithms to achieve heights of efficiency impossible for humans.
But why don’t we think this when talking to artificial intelligence chatbots? Most of them answer like dumb.
This is because we – humans – do not follow the pattern. A human user can ask a robot about the weather in Florida, then ask him how he can forget his ex and then a human can order a robot for pepperoni, but not as salty as the previous one. Was the bar.
what’s the problem?
While AI chatbots can analyze text, they do not understand it. Even the smartest of them are basing their responses on a keyword combo. People use hundreds of synonyms and thousands of word combinations of synonyms but different shades of the emotional spectrum. The lack of these feelings and colors is the main reason why machine learning chatbots seem stupid.
Another reason is even simpler. AI cannot know which word in our phrase is more important than others. For example, we want to order pizza. We ask our self-learning chatbot to add a double portion of cheese. Artificial Intelligence does not know if he should order pizza, if he is unable to take care of the double thing or he should not forget about the damn thing and please you with good pepperoni anyway.
AI is a best example of how the human voice processes Siri. It can book tickets, call, remind you about your plans. But it cannot solve problems or is more important about you.
The first successful attempt to solve the problem with an artificial interactive bot was taken in March 2016. Google introduced AlphaGo. It (he or she?) Can learn which of his actions helped him to achieve the goal. AlphaGo then makes conclusions for the future. And don’t worry, you’re not going to bring double-cheese pepperoni to check if you’re happy or not.
The modern-day AIS passes the Turing test and fools 90% of average humans and 30% of professional psychologists and linguists. David D. Laxton described his book “Artificial Intelligence in Behavioral and Mental Health Care”, in which AI simulated the behavior of paranoid schizophrenic and convinced 33% of experts that the results were generated by a sick human.
Modern-day neural networks have multilayer perceptrons. While typical low-level programming involves only 1 and 0, multiple cascades of semiconductor neurons allow varying the output signal strength between 1 and 0. which closely mimics how a human brain works. Does.
Working memory (we’ll talk about this a little later) is also an idea derived from a natural brain. It simulates how animals collect different experiences and develop certain behavioral patterns.
Alpha Go and the most advanced AI chatbot algorithms use so-called “deep learning” and “learning with reinforcement”. This technology is protecting people from stupid robots. Making the robot smarter, of course.
During the first phase, a layer of neurons receives the input signal. The second phase begins when input neurons distribute signals among other layers of semiconductor neurons. Such multi-layer cascades of neurons allow data to be processed in both linear and non-linear ways. It is the key to ultra-high speed data processing in deep neural networks.
Learning with reinforcement is derived from nature. Recall the puppies fed with sweetheart for every correct action. When the AI makes the right decision and gets the result, it gets some kind of reward. This is different from a commonly supervised machine learning because there are no pre-established results. AI must analyze its own results and find the optimal way to achieve a particular goal.