Source: Usa Pyon/Dreamstime.com
Takeaway: AI is a hot technology, but many people have misconceptions on what exactly it entails. Here we take a look at some of the myths surrounding AI and examine the facts.
Why is everybody talking about AI, yet we still do not see friendly robots like Data from “Star Trek” walking among humans? Did we remember to add RoboCop’s Second Prime Directive to their scripted patterns so they can “Protect the innocent” instead of exterminating humanity as soon as they gain full sentience?
Today, there’s a lot of confusion about what artificial intelligence (AI), machine learning and deep learning actually are, what “intelligent machines” can do, and what the current state of AI technologies actually is. It’s time to enjoy some good old debunking, so let’s bust the 10 most common myths about AI.
1. AI consists of intelligent robots or androids that look like humans.
Too much “Blade Runner” for everyone here, hmm? Although there’s a lot of general confusion between robotics and AI, they are two completely different science fields which serve different purposes. Robots are physical devices served by actuators and sensors to perform a wide range of tasks, such as building, carrying or dismantling products in factories.
2. AI, machine learning and deep learning are all the same thing.
Although they’re all parts of the same larger AI system, they’re three different things. Basically, machine learning is the method through which AI learns from external sources, as in using algorithms to discriminate data and determine its correct behaviors. Deep learning is just one possible technique used in practical applications of machine learning. It is based on neural networks (NNs) and is used to tell the AI what its probability is of making the right decision.
3. AI learns completely on its own.
Despite some exaggerated hype about AI that was allegedly able to learn on its own, it is still impossible to find an AI-powered system that has any real-world application that can grow from zero knowledge without human assistance. Any system that has to deal with hidden information or uncertainty of any kind cannot be “understood” by AI, which still needs to be fed input and data by humans. Also, every bit of information must have a clear purpose, something that AI cannot guess without external sources (not in the beginning, at least).
4. Chatbots are the most basic form of AI.
Again, even if there are some chatbots out there that make use of more or less rudimentary forms of AI, most of them are nothing but basic programs that interact with humans via text or voice interfaces. Rather than actually being “intelligent,” most chatbots have preprogrammed responses that are given in response to certain keywords in the user’s input. For a chatbot to become a true AI, it must possess several technologies that allow it to understand a human, learn about his or her needs, and react accordingly. It needs voice or text recognition software, sentiment analysis, some form of machine learning program and a natural language generation technology. (To learn more about chatbots, see We Asked IT Pros How Enterprises Will Use Chatbots in the Future. Here’s What They Said.)
5. The power needed to perform all future deep-learning operations is unsustainable.
It is undeniable that AI requires a lot of additional computing power to be trained and perform all its complex deep-learning operations. In a future where most enterprises will make use of AI to some extent, this problem may grow to epic proportions, making its use potentially unsustainable. However, AI may actually provide us with more power by staunching a perennial problem of energy production: power grids’ waste and inefficiency. Utility companies end up buying excess energy from private users, who also waste most of the excess electricity they generate since current grids were not built to accommodate the modern level of diversification. AI can come to our rescue by replacing old grids with newer, smart, AI-powered microgrids that know exactly how to distribute electricity in real time with the utmost efficiency.
6. It’s easy for an enterprise to rent the computing power needed to fuel AI operations.
… if AWS, Google, Microsoft and Alibaba Cloud weren’t currently centralizing the vast majority of the computing power available in the world. So AI developers currently have just two choices: renting it at exceptionally high prices or purchasing their own super-expensive hardware.
However, there’s a chance that this myth-debunking can be… debunked in the near future. A new company called Tatau developed a blockchain-based supercomputing platform that can solve the issue. Their solution allows the aggregation and reselling of the combined resources of a globally distributed network of GPU-based machines. Imagine cryptocurrency miners, gamers or other high-performance computers dedicating their compute power toward AI development. AI companies can tap into this underexploited source of GPU power to train their machine learning models at a much cheaper price. Note that this new platform may also provide an answer to the problem highlighted in point 5 since it promotes efficient use of currently untapped resources.
7. You need immense amounts of data to train AI.
Not necessarily. Sure, you need a lot of data and computing power to train an AI from scratch. And, albeit to a lesser extent, you need terabytes of data to train an AI to perform a complex task such as driving a car. However, depending on the field of application of the AI, pre-trained neural networks are flexible enough to be retrained only in some specific areas. The basic data framework may come from a larger, more general data set, with only the last part of the network needing to be replaced to “fill in the blanks” specific to that given use case.
8. AI will replace existing BI tools, making any previous technology obsolete.
That’s a bit of a stretch, to say the least. The majority of modern business intelligence (BI) solutions are highly scalable and often customizable, so that any future AI-based model can be easily integrated directly inside their platforms. Companies always prefer to implement only those solutions which come without any risk for workflow disruption, and AI technologies have adapted to this need. Therefore, most AI platforms are implemented via the web so no replacement is necessary or, in the worst-case scenario, can be safely implemented in phases.
9. Neural Networks are like biological networks but mechanical.
No neural network can even hope to reach a fraction of the complexity of the human brain. Despite many years of clinical and scientific research, we still fail to understand biological neural networks to their full extent since neurons fulfill so many different tasks with the human body (think about the difference between a sensory and a motor neuron) and even transmit information through many different pathways (using electricity, chemical potential and neurotransmitters). Neural networks can only understand very simple inputs in the typical 1 or 0 (“yes” or “no”) machine fashion. It’s like comparing the complexity of a military aircraft to a kite just because they can both fly.
10. AI will eventually become intelligent enough to understand that humans are dangerous to it and must be exterminated.
Well, we can’t actually debunk this myth since it’s not a myth. It’s a reality. Brace yourselves, because resistance is futile!
Jokes aside, simply put, AI has nowhere near the intelligence needed to understand the world around itself and make autonomous, rational decisions. Each algorithm is developed to perform one task and is not able to do anything outside that, let alone reach the ability to think independently. Computers use the “brute force” of their superior computational powers to find a solution to relatively simple issues, but they lack the understanding, perception depth, and strategic complexity to have a purpose outside the one they’re programmed for.
So rest easily, because AI is just going to be nothing but our artificial helpers and servants for a long, long time.