Digging Deeper | Is AI really a threat to humanity?

The movie showed the potential dangers of artificial intelligence when one cannot grasp, or even foresee, the extent of its abilities. While on that, also remember that this was movie is more than 50 years old! So we are not talking about all that new a phenomenon.

On a somewhat lighter note, for comicbook enthusiasts, there was a story of Mandrake the Magician, where a supercomputer named Goliath, which has been fed all the information in the world, goes rogue and starts building an army of robots in order to take over the world and make it perfect.

Cut to the present. What we have today are advanced versions of artificial intelligence that operate through little handheld devices. So much of our lives is run by some form of artificial intelligence, even though we may not be conscious of it. For example, a very basic form of artificial intelligence runs motion-sensitive appliances such as lights, fans and airconditioners. A more advanced operation that most of us use quite frequently is mobile navigation. Over the last few years, GPS and the ubiquity of data has made map directions and suggested routes so accurate and precise that an hour-long car journey can be predicted to a matter of a couple of minutes. Besides, the time is calibrated on a continuous basis as a result of continuous data crunching.

Or, take speech-to-text or personal assistants on your mobile phone. Within a short time, from the sample of a few hundred spoken words, they are able to understand the quirks of your accent and able to decipher your words. It was only half a dozen years ago that reliable speech-to-text required installing a massive application on your laptop and reading out to it a prepared page of text till the programme was satisfied!

There are hundreds of examples we could keep enumerating and explaining, but these few would give enough of an indication of how pervasive and integral artificial intelligence has become in our lives.

But these, experts would say, still operate in a very limited area and tackling highly specific tasks based on a given set of rules. All these operations essentially depend of massive data crunching applying the set of rules, leading to fresh calculations and fresh conclusions. We are still at some level talking about intellectual robotics, if one might call it that. An advanced level of this would be the artificial intelligence that enables the fly-by-wire function in airplanes. Here, dozens of variables are being continuously monitored, based on which minute course corrections are made continuously.

Artificial intelligence has been in increasing use in the area of gaming. Take the game of chess. Deep Blue was the first computer that was developed to play chess. The project was done at Carnegie Mellon University in 1985, and was later taken over by IBM. Deep Blue created history by becoming the first machine to win a game against a reigning chess champion, though it took a decade of refining for that to happen. Remember that the opponent was Garry Kasparov. However, that first win was followed by three losses in the six-game match. A year later, however, an upgraded Deep Blue beat Kasparov in another match.

Again, Deep Blue, and many later chess playing programs, relied on the same method of massive data inputs, enormous processing power and brute force computing.

This is where Google’s DeepMind and AlphaZero need to come in. A year and a half ago, AlphaZero, the game-playing artificial intelligence, beat the world’s best chess program. The notable point is not that, though. It is that AlphaZero was the first truly successful artificial intelligence game program.

In its earlier avatar as AlphaGo, a program designed to play the Chinese game Go, had beaten Ke Jie, the best player in the world, twice in succession, in an exhibition of intelligence that Jie referred to as “godlike”.

AlphaGo was later tweaked to be able to learn and play other games, and christened AlphaZero. And in an awesome – some might say frightening – display of learning, AlphaZero, which had been fed with nothing but the basic rules of chess, beat Stockfish 8, the most powerful chess playing program. And it did this after teaching itself to play for just four hours!

Even though this still falls in the realm of specific tasks within given sets of rules, we can expect that this is what the future of artificial intelligence is going to be like. It will surely go beyond being good at chess to being good at learning and being good at most other things that a human being can do. But, how good can AI get at tasks that require general intelligence, where the parameters are unknown and continuously changing? How well can it operate in situations that require fuzzy logic? And if AI does get good at that kind of learning and application, how could it be directed to do only those things that benefit humans and stopped from going beyond or aside?

Benefits of AI

First things first. It is clear that artificial intelligence has enormous potential in expanding the range of human knowledge and endeavour.

Broadly speaking, artificial intelligence is already making distinct differences in convenience and comfort, to say nothing of information. All three are served usefully by the various digital personal assistants that a lot of services and products come with. When coupled with the advances in the area of Internet of Things (IoT), we reach a point where digital assistants not just follow instructions, but operate also as a perfect butler, reminding you of things, cleaning up after you, and preceding your arrival by keeping your house and appliances in exactly the condition you prefer. Assistants have become personal and efficient to the extent that one might even form deep emotional bonds with these products and software!

Too often, when we chat online, we are unable to tell whether the voice or text writer at the other end is a human or a bot – that’s how good the software is at effective communication.

When automation extends to where the machines and the programs running them can make decisions on their own, we have a situation where we save time and money, and gain speed and accuracy. The amount of data that is available, when it is fed into a computer system, is able to enormously help in the diagnosis of diseases. For instance, the various little deviations from normal can be computed over time, with algorithms dealing with multiple variables to the extent that it can provide pointers to impending diseases or physiological conditions much before the symptoms manifest fully enough for a definitive diagnosis. And we are talking here of things like cancer, diabetes, dementia. The gains here do not have to be explicitly stated to be understood.

Extend this data processing and analysis, and we could have a situation where these intelligent programs would be able to play an enormous positive impact worldwide in the areas of disease prevention, food sufficiency, water distribution, and efficiencies in all areas of human need and endeavour. We are also at the stage where neural networks are trying to recreate the workings of the brain. This could provide crucial insight into how our brain functions, and perhaps give us a peek into the mysteries of life itself. To quote Stephen Hawking, “The potential benefits are huge; everything that civilization has to offer is a product of human intelligence; we cannot predict what we might achieve when this intelligence is magnified by the tools that AI may provide, but the eradication of war, disease, and poverty would be high on anyone’s list. Success in creating AI would be the biggest event in human history.”

Hawking then proceeded to say that there are no real limits to what can be achieved. As he put it, “There is not physical law precluding particles from being organized in ways that perform even more advanced computations than the arrangements of particles in human brains.”

Threats posed by AI

So, it looks like machines will be better than humans in many ways. So what might that mean in ways that could be unpleasant for humans?

The same Hawking has also commented upon the disturbing aspects of artificial intelligence and how we might understand – or, not understand it. In his own words, expressing disapproval about the manner in which AI was being dealt with, “Facing possible futures of incalculable benefits and risks, the experts are surely doing everything possible to ensure the best outcome, right? Wrong! If a superior alien civilization sent us a message saying, ‘We’ll arrive in a few decades,’ would we just reply, ‘OK, call us when you get here, we’ll leave the lights on?’ Probably not, but this is more or less what is happening with AI.”

He was talking about the existentialist risk to humanity from super-intelligent machines and programs, about the idea that they could supercede humanity and pave the way for the ultimate extinction of the species. Elon Musk, entrepreneur who has also been active on the cutting edge of technology, has called Artificial Intelligence “more dangerous than nuclear warheads.” To these two important voices, Bill Gates has added his bemused expression: “I don’t understand why some people are not concerned.”

But there are voices that rubbish these fears too. Jaron Lanier, computer philosophy writer and computer scientist, even dismissed the talk about machines even being intelligent in any way.

While the extent to which Artificial Intelligence might progress as a possible threat to humanity itself is in the middle of debate, there are many aspects of AI that have been flagged as potential dangers if there is no attempt to understand the implications of the sweeping developments happening in the field.

One of the threats most talked about is how AI might behave when entire operations are left to it to be done. If AI is taken to its logical conclusion, it must reach a stage where it would react and respond in the way humans would. In which case, one would expect it to take decisions that work in its favour. How does one manage AI so that it works not to its own best benefit but to that of humans?

Another threat or area of discomfort deals with the algorithms themselves. They process humongous amounts of data to work efficiently and this automatically leads to the problem of privacy. This also results in personal profiling that includes aggregation of almost every bit of personal detail that results from one’s online activity. What is the level of comfort, and how far is this profiling likely to go? Already, we have a situation such as in China, where profiling and facial recognition has led to almost complete monitoring of citizens, their status and their activities. This affects a person’s life from being unable to access products and services to being branded in undesirable ways on the basis of temporary or questionable data.

With algorithms, there are other concerns such as the lack of transparency of data, which means that we can be in a situation where we don’t know how data has been processed to arrive at a conclusion or decision. This would be crucial in medical or military decisions. This is especially true of weak, or narrow, intelligence systems, which reach decisions mainly on the basis of data crunching. If, along the way, one step goes awry or involves a non-optimal call, it could result in disastrously unpredictable results. This could happen in, say, in utilities supply chains or decisions based on reading economic data by artificial intelligence.

Unlike narrow AI, general AI, also sometimes referred to as strong AI or full AI, is what approximates to human intelligence. Two years ago, about 40 organizations were working on the concept of General AI. Without set guidelines, research could reach undesirable points.

Social manipulation using powerful algorithms is already a reality at various levels. This has the ability and power to change and direct human behaviour that can skew important events around the globe. Election results being affected through social media manipulation has already been indicated, with investigations being conducted on how and how much this might have happened. Here, we are also talking about targeted news, fake news, audio and video manipulation, where it becomes difficult to distinguish between fact and fiction.

Of course, some of the extreme prognostications regarding the threats of Artificial Intelligence have also been pooh-poohed. However, the very fact that there are serious threats that continue to be seriously discussed as possibilities must itself make us ask questions and tread with caution in how we handle this important step in the history of mankind. There is the general consensus that AI is here to stay and is an inevitable part of the future of mankind. It is truly up to us humans to know where this might lead and to see that the enormous benefits and promise of Artificial Intelligence are used for the benefit of the whole of humanity and not for the short term benefit of a few with a narrow profit motive. We have humanity at stake.