What is General Purpose Artificial Intelligence?

Artificial Intelligence (AI) is the hottest industry today. Billions of dollars are being spent worldwide to research and develop new technologies in the field of AI. One of the biggest goals of these researches is to develop General Purpose Artificial Intelligence (GPAI). What is GPAI?

We already know AI is the ability of computing software programs to learn as to how to make decisions or do things based on complex input data. For example, suppose you are thinking about investing your money so that your investment is risk-free and at the same time provides a good return on your investment. We know that there are many financial instruments available in the market which; vary considerably on the parameters of risks and returns. Investing money in shares of companies can provide good returns but are highly risky. You can lose money if share prices fall. But you can get good returns if share prices rise. At the other extreme are fixed deposits of banks. Returns on investment here are totally risk-free but returns are very low. Then there are many financial instruments like mutual funds which are less risky but can provide more returns compared to fixed deposits.

People invest their money after thinking about these factors. In a way, we can say that we use our intelligence to make a decision about an investment. AI software programs have been developed to do such tasks even better than us human beings. In fact, many AI software programs have been developed to do many such tasks and are already in the market. These AI software programs are really smart and can do things better than human beings. But if these AI software programs have better intelligence as compared to human beings?

Even though these AI software programs can outsmart human beings in performing some specialized tasks, nevertheless a human being has far more intelligence as compared to these AI software programs. A human being is capable of doing many things even when not much information is available or no prior experience exists. AI algorithms, on the other hand, are capable of doing only a specific kind of tasks. For example, a human being can easily perform language-related tasks (reading, writing, speaking, listening etc.), physical tasks (walking, manual work etc.), mental tasks (financial investment, marketing strategy) etc. To perform all these tasks, we may need different AI software programs. We still do not have an AI software program which can perform all these different kind of tasks alone.

What is General Purpose Artificial Intelligence (GPAI)? It is the ability of an AI software program to perform a different kind of tasks just like a human does. In technical terms, it means that we have to build an AI software program which can be trained to perform a different kind of tasks. When it happens then such an AI software product is termed as GPAI.

Recently DeepMind Technologies (an AI company owned by Alphabet Inc.) has announced that it has achieved GPAI with their AI software program known as AlphaGo. This AI software program can be trained to play many kinds of computer-based games like Go, Chess, Minesweeper etc. This means just one AI software product can perform many types of tasks after learning all those tasks. This indeed is a great milestone in the field of AI!

But AlphaGo still has its limitations. Foremost limitation is that it works in an idea environment where it receives signals just from the game which it is currently playing.  There is no “noise” (no distractions of the real world where too many things happen where we have no control). The other limitation is that it works only at the visual cognition level. Humans interpret their environment from signals related to visual, sound, smell, touch and taste.

AI research and development is still to go a long way to build an AI software program which can have the same level of intelligence as compared to a human being. The AlphaGo AI software program still has not achieved the same level of intelligence as compared to a human being. But in the next 10 years or so; we will be able to see AI software programs which may achieve or even surpass human-level intelligence!