Over the past five years, artificial intelligence (AI) has again been spoken about in a consistent way and not because of the cinematic work of the same name.
The increasing interest from companies (both developers of the technology and those who benefit from it) and the international press has undoubtedly been brought about by the enormous technological advances that increase computing capacity (ever more available with as-a-service models in the cloud), the availability of large amounts of increasingly sophisticated data and tools for its analysis which have accelerated the development of artificial intelligence itself.
The topic is red hot and, especially in the last two years, much has been written about it but many companies often view the matter with a certain level of confusion and it is not always clear what exactly is being said. Speaking in general terms, we can describe AI as a scientific subject, a discipline that studies methods and tools that can solve problems or perform particular activities of human intelligence.
It is quite clear that this is a very broad description that encompasses different abilities, from reasoning to behavior, difficult to confine in unique definitions and models. And it is for this reason that there is not yet a standard for defining artificial intelligence.
If we want to attempt to compare human capabilities with artificial ones, we should be able to say that AI is able to:
- act like a human being;
- think in a “human” way, that is, using cognitive functions and using logic (rational thinking);
- solve problems like a human, arriving at the result based on the information available.
All in all, if we go no further than this initial analysis, we may also come to think that AI has now reached its full maturity, but this is not so.
A bit of clarity between weak AI and strong AI, machine and deep learning
That which is certainly firm and clear is the separation between weak and strong artificial intelligence:
- Weak AI: these are systems that can simulate certain cognitive functions of humans, in particular logical reasoning for solving problems or making decisions (activities behind which we find machine learning technologies);
- Strong AI: “wise” systems with an autonomous sense of reasoning and thinking (functionality that “machines” reach manually, independent of humans, thanks to deep learning technology).
This distinction is important because it depends on the substantial differences between AI, machine learning, and deep learning:
Artificial Intelligence = the capability of machines to perform some typical tasks of human intelligence (planning, understanding of language, image and sound recognition, problem solving), distinguishable in weak AI and strong AI;
Machine Learning = born as an algorithm to improve artificial intelligence, it is the system that allows machines to learn to the point of performing a task or activity (artificial intelligence, therefore) without this activity being programmed.
Machine Learning is a system that “trains” artificial intelligence to do a task by correcting, adapting and improving this ability through learning). In other words, these are algorithms that can learn from data and make predictions through data, with some intertwining of disciplines such as statistics and predictive analysis.
Making the difference in these systems is the “learning model” of the algorithms:
- with didactic supervision: there are both input and output examples and the algorithm uses this data to learn how to behave;
- without didactic supervision: in this case the algorithms find their own models without particular input or output data but by “simply” mapping the results;
- reinforcement learning: the system learns through a sort of award coupled with the correct results (the more it gets rewarded, the more it improves its capabilities).
Deep Learning = the ultimate frontier of artificial intelligence, the most advanced part of machine learning (which is a subcategory) because it’s inspired by the structure and functioning of the human mind. While machine learning is the algorithm that “trains” artificial intelligence, deep learning is the algorithm that leads AI to emulate the human mind. From a technical point of view, deep learning requires the use of deep artificial neural networks, algorithms and a computational capacity to “replicate” the human brain.
This is the case of “high-power” algorithms because their learning model is based on different layers of computation and analysis. Each layer uses inputs from the previous layer and provides output to the next one. This model lends itself to the analysis of large data bases and is already in use, for example, for pattern recognition, speech recognition, image recognition and natural language processing.
The combination of neural networks (available via clouds), machine learning technologies, and the myriad of data available through the Internet have made deep learning one of the most interesting and exciting subcategories of AI. We were all very amazed to see Google’s DeepMind beat the (human) Chinese champion at Go!
Cognitive systems: the artificial intelligence market for companies
Based on the average growth estimates of AI Enterprise’s market (i.e. for business solutions) compiled by large analytical companies (McKinsey, Frost & Sullivan, EC, Forrester, Idc, Gartner), within the next decade investment in artificial intelligence technology will go from the current $200 million to over $60 billion.
When it comes to artificial intelligence in the business world, the term cognitive systems is often used to identify technological and functional aspects whose purpose is not to replace man but to increase their capabilities, especially in 4 functional areas (which are also the areas where technologies are developing the most):
- Understand: technological systems that deal with the correlation of data and events for the recognition of texts, images, tables, videos, and so on. And the analysis needed to extrapolate information useful to the decision-making process (it is undoubtedly the area where the highest qualitative performance is already present);
- Reasoning: the so-called “reasoning” technologies, those that allow aggregation of information by means of algorithms that simulate the neural connections of the human brain, for example to make decisions or solve complex problems (there is still much to improve in this area);
- Learning: here we talk about real algorithms, i.e. technologies and systems that will allow more and more sophisticated cognitive systems (artificial intelligence) to return more and more effective and high quality outputs with respect to the initial data;
- Human Machine Interaction: in this case we are talking about the technologies of communication/relations based on natural language that will facilitate and simplify the interaction between man and machines (such as, for instance, with voice-assistants or chatbots).
A technology map to understand “what to buy”
The AI technological ecosystem for the business world is already quite rich but the maturity levels for solutions are varied. Before investing, a company should therefore understand what today’s market offers, what technologies have really matured and can generate a business value and which ones are those worth watching because they’re promising but still evolving.
By gathering together several studies and analyses (Frost & Sullivan, Harvard Business Review, Stanford University), Callaghan Innovation has translated into a graphical map the “technology readiness” of artificial intelligence based on both the time span of development and as being “field operational “, i.e. the scope where certain technologies can best express their potential.
That of self-driven vehicles is perhaps the best-known area; interest has increased dramatically thanks to drones but, Tesla and Google being complicit, we have moved very quickly also regarding more traditional vehicles (from Bmw to Audi, Ford and Toyota) which already incorporate “driving assistance” that uses artificial intelligence algorithms.
Chatbot, pattern recognition and the use of natural language (e.g. through apps and services such as Siri, Cortana, and Alexa) are other areas where we can say that a good level of technological maturity has been reached.
However, it is best not to confuse chatbots and voice assistants with technologies such as those used in virtual assistants or social assistance/support robots; in this second case, data analysis, self-learning and the recognition and use of natural language are not enough. They need superior skills more pertinent to the emotional and social sphere, fields where technology has yet to mature.
Looking to the future, one of the most promising evolutions, but one which will still take a few years of research and experimentation, is that of real time emotion analytics, artificial intelligence capable of analysing and interpreting human brain signals through tone of voice or facial expressions, thus recognizing emotions.
Over the next two to four years we will have solutions that will raise the bar of customer service even higher, in terms of value.
