As the buzz that surrounds robots with Artificial Intelligence (AI) increases, our expectations become sharper. Naturally, it makes most of us intensively question the perspectives of the field.
The problem is that Artificial Neural Networks (ANN), which has come to be the main advancement in AI, have their black holes that don’t allow scientists to track the way they make predictions. On the other hand, we still don’t have robots with Artificial Intelligence that really understand what they hear or see.
The whole picture looks like, despite all the obvious advancements, there’s still nothing definitive about the AI revolution in any observable future. So, is there any real ground for us to expect the rise of robots with Artificial Intelligence in the first decade of 21st century? I invite you to answer this question together.
Future for robots with Artificial Intelligence: to be or not to be?
Before we actually get our hands on drawing the future of Artificial Intelligence, let’s think for a bit about the whole picture. The latter is just impossible without the ability to clearly see the particular shortcomings of Artificial Intelligence that scientists have to overcome. That’s what we are going to start with.
Current AI problems
1. The measured computing power
In 1965, co-founder of Intel Gordon Moore (learn more about him on Forbes) predicted doubling of the number of transistors every 12 months, which, in other words, means that the computing power increased dramatically for the half of the century. Later, this prediction was called “Moore’s law”.
Such interconnected fields like Machine Learning (ML) and Data Science (DS) owe their existence to the increase of computing power, which has reached the point where massive data sets can be effectively processed by the computer “brain”.
The problem is that Moore’s law wasn’t going to last forever. Moreover, the growth of computing power has already slowed down and will completely expire in five years. This may not contain anything optimistic for Computer Science unless the alternatives already exist.
The possible solution: supercomputers with the architecture that differs from the conventional one (for instance, brain-like architecture).
2. The opacity of ANNs
We love secrets. Most of the people actually do. This is something in our human nature that would prefer to leave a place for unknown – something that is yet to be discovered. However, when it comes to trusting Artificial Intelligence, you would, most likely, want to know how exactly it works.
ANNs are the most advanced and progressive techniques in Machine Learning. All the robots with Artificial Intelligence that surprise you by their “smartness” are, indeed, based on the complex ANN. This is the problem: the more sophisticated the network is, the more opaque it is.
ANNs can make predictions about just everything they are pre-programmed for. All too often, to get a highly precise prediction, one needs to consider a lot of data. Mapping all the dots to each other inside the data model ANN is based on makes the network more complicated.
Finally, you get the prediction of the high accuracy but you don’t understand how that complex network produced it. In Deep Patient, scientists trained neural network on the thousands of health records. What they finally got was the tool that can predict numerous diseases – from heart failure to cancer to diabetes – which doctors are afraid to use due to the opacity of the process.
Although for such fields like customer service, where you just chat with the robot about general issues, this lack of transparency can hardly cause any substantial problems, the fields that need it at most (like Healthcare) wouldn’t want to blindly rely on the results.
The possible solution: letting ANNs follow the natural path of evolution (where – just like with the human brain – not everything still can be explained) or create networks with the self-explanatory possibilities.
3. Lack of memory
Memory is an old problem of AI. When you train any network to be able to predict something particular, it takes data and time. After the training is over, you can deploy the network for solving real-world problems.
However, if you decide that you need to extend the range of issues it solves, you will need to do everything from the beginning. This is all due to the lack of memory. Unlike to us, humans, whose ability to master new skills is deeply rooted in memorizing past experiences, modern robots with Artificial Intelligence are pretty measured.
The possible solution: recurrent neural networks with the associative memory (based on storing patterns and retrieving them in order to match with the input).
4. No real understanding
This is, perhaps, the main problem of AI, which consists of all other problems or, at least, is tightly connected with them. Even though, we say “Artificial Intelligence”, in fact, so-called “intelligence” that is present in some of the programs is very far from being human-like.
Despite all the image, speech, and voice recognition, smart text analytics, and other skills of smart systems, even the most skillful of the robots with Artificial Intelligence don’t understand what they analyze. ML techniques make possible to teach computer programs to recognize patterns or figure them out but they never get to see the whole picture.
Every time you get to chat with robot – even with the most intelligent one – there’s the point when you realize that it’s a virtual agent, not a human being. It happens because AI part behind the agent is trained to recognize only certain parts of speech but can’t grasp the underlying concept of what you’re saying.
The possible solution: perhaps, the answer lies in the ANNs that need to develop enough to gain consciousness or maybe the approach that will lead to conscious machines is somewhere beyond the ML field.
Where do current AI problems leave us?
Peter Stone, a computer scientist at the University of Texas, is convinced that AI is about to shape all the sides of our life by 2030. He says: “We believe specialized AI applications will become both increasingly common and more useful by 2030, improving our economy and quality of life. But this technology will also create profound challenges, affecting jobs and incomes and other issues that we should begin addressing now to ensure that the benefits of AI are broadly shared.”
Peter Stone chairs the AI100 Standing Committee of the massive research project “One Hundred Year Study on Artificial Intelligence”. Every five years AI100 project will come up with the detailed report on the status of AI technologies, and the first one was already released in 2016.
According to the first report of AI100 project, by 2030 we will have:
– autonomous transportation and, as the consequence, a new urban organization;
– enhanced robots’ services and their interactions with people driven by Artificial Intelligence for Robotics (“robots will deliver packages, clean offices, and enhance security”);
– the close cooperation of AI systems with healthcare providers and patients;
– meaningful integration of interactive machine tutors with face-to-face learning;
-”improved cameras and drones for surveillance, algorithms to detect financial fraud, and predictive policing”;
-task replacements (“AI will likely replace tasks rather than jobs in the near term, and will also create new kinds of jobs”).
There’s plenty of things to do before we get the real human-like intelligence around us. Despite all the difficulties we are facing, the perspectives are very optimistic for every field (education, healthcare, public safety and security, employment, transportation, household services, etc.)
If you’re interested in modern Artificial Intelligence for Robotics, the easiest way to give it a try is to get involved in the chat with robots, which are widely available on the Web. You can learn more about chatbots here or order the one for your business here.