What is AI
General introduction to what is AI.
What exactly is AI?
Fundamentally, Artificial Intelligence (AI) represents a field within computer science that focuses on enabling machines to discover solutions to complex problems, in a manner that’s a bit like how humans think. Typically, this requires adopting certain characteristics of human intelligence and implementing them as algorithms in a manner suitable for computers. AI constitutes the domain of computer science devoted to the automation of intelligent actions.
Understanding and generating natural language, recognizing patterns, making decisions based on data, and even learning from experience are some tasks that AI enables machines perform by mimicking human cognitive functions. At present, visual perception, speech recognition, decision-making, and language translation are some important functions of AI.
Take a few minutes to think about where do we encounter AI in our everyday lives.
Virtual assistants like Siri, Alexa and Google Assistant rely on AI to understand and respond to our voice commands.
Social media platforms, online shopping platforms and streaming services make use of AI to recommend content and products to users based on their preferences and user history.
Moreover, navigation apps, medical diagnoses, autonomous vehicles and so on are a few other examples of AI powered domains.
Given the various interpretations and applications associated with AI throughout its history, it is difficult to provide it with a precise definition. what we can do is, to derive a comprehensive idea of what AI involves with the help of a few historical definitions and examples – as any precise definition of AI becomes outdated as the technology advances further in to the future.
let's examine a few attempts to define Artificial Intelligence by experts throughout history.
Definition 1
A good starting point for an introduction to the term “AI” is the 1956 Dartmouth summer research project on artificial intelligence, where the term was coined by American computer scientist John McCarthy and collaborators, who is now known as one of the founding fathers of AI. He used the term to describe the new discipline of thinking machines. His definition of AI provides a foundational understanding of the field. He defines it as follows,
It is the science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computers to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable.
McCarthy’s definition points to a few important things about the primary nature of AI.
it points to how AI involves both theoretical research and practical implementation. Suppose you want to create a Chess-playing AI, the process will involve studying the game of chess and developing the algorithm to enable a computer program to play the game.
it suggests that the goal of AI is the creation of machines or computer programs that can make intelligent decisions – specifically, a program that would make decisions similar to how humans would make decisions. In the case of the chess-playing AI this would involve the computer program’s ability to come up with strategic moves to win the game.
another important aspect of AI implied by McCarthy’s definition is that- even though the function of AI is to understand human Intelligence, its operation is not limited to biologically observable methods. This statement indicates that an Artificially Intelligent system doesn't necessarily need to replicate human intelligence but can be inspired by it. It can employ different, non-biological methods to exhibit intelligent behaviour, such as its own algorithms and techniques.
A chess-playing AI program won’t think like a human. but it will demonstrate human-like intelligence using algorithms and techniques.
McCarthy’s definition was seminal to the understanding of AI during its earlier stages. His definition of AI implies a focus on replicating or mimicking human intelligence. At present the scope of AI has evolved to incorporate a wide range of tasks and applications that do not necessarily resemble human intelligence. For example, modern AI technologies are excellent in tasks which are not related to human like thinking, such as data analysis, optimization, and automation.
Definition 2
let’s take a look at another significant definition of Artificial Intelligence.
Computer scientists Stuart J. Russel and Peter Norvig, who authored the book Artificial Intelligence: A Modern Approach (1995), define AI as;
the study of [intelligent] agents that receive precepts from the environment and take actions.
This definition implies that,
AI is concerned with creating and studying agents, which can be software programs, robots, or any other entity capable of interacting with its environment.
The agent receives information from its environment, obtained through sensors or any other methods of perception.
Take the example of a self-driving car as such an artificially intelligent agent. It receives precepts or information of its environment from various sensors like camera, radar, GPS and so on, which it utilises to take strategic actions.
The agent takes actions and reacts in response to the information received from its environment, to achieve its objectives. In the case of the self-driving car, the sensors provide information about the car's surroundings, such as the positions of other vehicles, pedestrians, traffic lights, road signs, and road conditions. The car then utilises this information to take decisive actions.
The function of AI involves defining a set of rules which govern how an agent should respond based on the precepts it receives. The self-driving car's AI system utilizes a sophisticated algorithm to analyse data from its sensors and decide on the right course of action. For example, when the car's cameras identify a red traffic light, the AI system interprets this data as a signal to stop the vehicle.
Therefore, according to this definition an agent is intelligent when:
Its actions align with its situation and goals
it adapts to different surroundings and objectives which are subject to change,
it gains knowledge through practice
it makes suitable decisions within its perceptual and computational limits
Russel and Norvig, further elaborates the definition of AI by identifying four schools of thought for AI. Which are as follows:
Systems that think humanly
Systems that act humanly
Systems that think rationally
Systems that act rationally
systems that think humanly
This school of thought centres on designing AI systems that imitate human cognitive functions and thought processes, with an emphasis on comprehending and duplicating human patterns of thinking and intelligence on a machine.
Example: GPS (general problem solver) was an early computer program that attempted to model human thinking. The developers were not much interested in whether or not GPS solved problems correctly. They were more interested in showing that it solved problems like people, going through the same steps and taking around the same amount of time to perform those steps.
systems that act humanly
Some researchers focus on creating machines that act like humans. That is, systems which can perform tasks and act in a way that appears human-like. It emphasizes on outward behaviour and performance rather than internal cognitive processes. The first proposal for success in building a program that acts humanly was the Turing Test. To be considered intelligent a program must be able to act sufficiently like a human to fool an interrogator. In order to pass this test, a program would require to have natural language processing, knowledge representation, automated reasoning, and machine learning.
systems that think rationally
the third school of thought focuses on developing machines that think rationally. Such an AI system is capable of logical reasoning and decision-making founded on formal rules and the representation of knowledge. It places formal logic and mathematical reasoning at the core of AI development.
Example: In the field of medicine, there exist expert systems that utilize a predefined set of rules and logical reasoning to make diagnoses by analysing patient information and symptoms.
systems that act rationally
the fourth school focuses on developing AI systems that make decisions and take actions to achieve specific goals or objectives in a rational and logical manner. It doesn't always imitate human behaviour but strives to achieve intelligent results. As the most recent perspective in AI, this school attempts to define the concept of rational agents. As explained by Russel and Norvig, an agent is essentially something that takes action, and a rational agent is one that acts in a manner to attain the optimal outcome or in situations of ambiguity, the best possible outcome. (Russel and Norvig, 2016)
Example: Chess-playing programs such as IBM's Deep Blue utilize a search algorithm to make choices that lead to the best moves and victory in the game, despite not employing human-like thought processes.
Definition 3
Elaine Rich in her book Artificial Intelligence (1983) provides a concise and evergreen definition of AI which is as follows:
Artificial Intelligence is the study of how to make computers do things at which, at the moment, people are better.
This definition highlights that the main goal of Artificial Intelligence (AI) is to create and develop computer systems that can do tasks or exhibit intelligence in areas where humans are typically really good. For instance, one challenging task that humans excel at is understanding and using language in all its complexities, including context and emotions, in both written and spoken forms.
Natural Language Processing (NLP) is a field in AI that aims to make Chatbots and virtual assistants that can understand and generate human language in a smart manner. Researchers and engineers who work on developing algorithms for NLP use machine learning techniques to teach computers to study and reproduce text and speech in a human-like manner. We will be discussing more about NLP and Machine Learning in the upcoming lectures.
While computer programs have made progress in things like understanding subtitles, humour, and context-specific conversations, they still have a long way to go to match human-level language skills.
Even though Chatbots can provide helpful answers to common questions and have simple straightforward conversations, they may still struggle in the case of emotionally charged or context-specific interactions that humans are able to handle effortlessly. Making computers perform as well as humans in this area is an ongoing challenge, and it underscores the core of AI, which is about teaching computers to do tasks where humans are currently better.
Definition 4
In a scholarly paper titled What is AI, Anyway? (1987) American AI theorist and cognitive scientist Roger C. Schank, delves into the scientific and technological objectives of artificial intelligence (AI) and addresses the core challenges within AI research. His definition of AI is as follows:
AI is the science of endowing programs with the ability to change themselves for the better as a result of their own experiences.
Schank makes a few important observations on the nature and fundamental aspects of AI. A few them are as follows.
He addresses how AI is most often misunderstood by people from other fields and that at times even AI practitioners are uncertain about what AI stands for.
Schank acknowledges how AI encompasses different things for people from different disciplines such as mathematics, software engineering, linguistics and psychology.
AI researchers emphasize the importance of using established mathematical formalisms for effective knowledge representation in AI programs. They view the quest for suitable formalisms as a central aspect of AI research. Therefore, for these researchers AI is mathematics.
Many AI researchers believe that building grammatical structures of languages like English on computers qualifies as AI. This perspective conflicts with traditional linguistics, which didn't initially associate with AI. As funding shifted from linguistics to AI, some linguistic theories not originally designed for computers are now being presented as AI models.
Schank also questions whether AI is similar to psychology as some AI researchers prioritize building a comprehensive computer model of human cognitive processes. While many AI researchers are indifferent to understanding the human mind, it remains the only form of intelligence we can reasonably study. However, this approach raises concerns about computer models displaying intelligence without mirroring human functioning, leading to a focus on the broader nature of intelligence beyond its human embodiment.
According to Schank, the question of what AI is, cannot be fixed with a single definitive answer. It is heavily dependent on the goals of the researchers and the methods by which they develop the AI model. Moreover, it’s about results.
Schank also points out that the debates on what AI is, exist solely because AI’s development remains unfinished and may only arrive at a conclusion once AI reaches the levels of advancement envisioned by authors of science fiction.
According to Schank AI as a discipline has two central goals, which most AI practitioners agree upon. According to him,
“The primary goal (of AI) is to build an intelligent machine. The second goal is to find out about the nature of intelligence. Both goals have at their heart a need to define intelligence.”
let’s take a closer look at this emphasis.
there are two radically different ways machines powered by AI might produce outputs resembling those of humans. First, machines could be programmed to model or mimic human cognitive functioning. For example, AI programs can solve many problems using strategies closely resembling those used by humans. A major goal of this approach is to increase our understanding of the human mind.
Second, machines could simply be programmed to perform complex tasks (and easy ones, too) totally ignoring the cognitive processes humans would use. The chess computer Deep Blue that beat Garry Kasparov exemplifies this approach. It had fantastic computing power, evaluating up to 200 million chess positions per second. Thus, Deep Blue’s huge advantage was fantastic processing speed rather than the cognitive complexity of its operations.
When we are trying to approach AI from a philosophical perspective, there are a few important questions that we have to consider. Such as, what is intelligence? How can one measure intelligence? What is it to have a mind? How does the brain work? How do humans actually manage to behave intelligently? And so on.
In their proposal for the Dartmouth project McCarthy and his collaborators suggest that machines can be made to mimic “every aspect of learning or any other feature of intelligence”. McCarthy identifies features of intelligence as the use of language, the formation of abstractions and concepts, solving problems now reserved for humans, and self-improvement.
This points to the first problem in understanding AI, namely its aim to replicate or emulate intelligence.
Intelligence itself is a contested concept and it is not clear which form of intelligence AI would have to replicate, in order to be worthy of being called AI. Biological organisms, including humans, operates on different principles from digital technologies. Humans have access to “mental abilities, perceptions, intuition, emotions, and even spirituality”. Should AI be replicating all of those?
This, in turn, points to the second problem in understanding AI. Are there barriers that AI, as a digital technology, cannot overcome, aspects of intelligence that cannot be digitally replicated? This is an interesting question that has been debated for a long time. It is ethically significant because it has a bearing on whether AI could ever be considered an ethical subject, i.e., whether it could have moral obligations in itself.
Both problems of understanding AI – namely, what is human intelligence and which part of it might be replicable by AI – make it difficult to define AI. The conceptual nuances of AI have led to a situation where there are many competing definitions covering various aspects.
in simple terms, Intelligence is the ability to behave adaptively and to solve novel problems. most importantly, intelligence is a general ability that is displayed with respect to numerous very dissimilar new problems rather than being limited to problems of a single type (e.g., problems in mathematics)
Among AI practitioners there exists conflicting opinions on what constitutes intelligence. AI experts generally share a common understanding of what AI is and what its goals should be. According to Schank, to solve this problem, one approach is to make a list of qualities we think an intelligent being should have. None of these qualities alone would define intelligence. In fact, something could be clever even if it doesn't have all these qualities. But each quality is an important part of intelligence in its own way. The five features that he considers important are as follows:
Communication: an intelligent entity can be communicated with. According to Schank, “the easier it is to communicate with an entity, the more intelligent it seems.”
Internal knowledge: intelligent entities are expected to know about themselves. Even though it is hard to examine the internal knowledge of an intelligent agent, the best way to do it is to ask and observe. If we get satisfying answers, we can assume that the entity has some level of intelligence.
World knowledge: awareness of the outside world and being able to use the information you have about it. This also means having the ability to remember past experiences that were stored in a way that makes them useful in different situations.
Intentionality: which means having a goal driven behaviour. Being aware of what you want and having the framework to achieve it.
Creativity: which is the ability to see things differently. Ability to adapt to your surroundings and learn from experience.
It is important to understand that AI is primarily the result of humanity’s fascination with Intelligence. Or rather, the nature of intelligence, the strengths and limitations of the human mind. This fascination persists across all of human history because of the fact that human mind is so complex that no one has been able to fully comprehend its nature and capacities.
The evolution of AI has witnessed a significant transition from attempting to grasp the intricacies of human intelligence to tackling intricate problem-solving tasks, such as medical image diagnosis. What prompted this shift? A pivotal factor lies in the immense intricacy of human intelligence, including the human brain, making the former approach exceptionally challenging to implement effectively. Conversely, the latter approach has shown some promise. However, it's important to note that despite this shift, numerous AI experts persist in their pursuit of the ultimate objective, which is attaining artificial general intelligence.
Conclusions
Artificial Intelligence (AI) is a field in computer science that aims to make machines solve complex problems in a human-like way by borrowing human characteristics and converting them into algorithms.
AI is widely applied in virtual assistants, social media recommendations, navigation systems, medical diagnoses, and autonomous vehicles.
AI encompasses a wide array of subfields, ranging from general-purpose tasks like perception and logical reasoning to specific applications such as chess playing, mathematical theorem proving, poetry writing, and disease diagnosis.
Scientists from various domains often transition into AI, where they find the tools and vocabulary to systematize and automate intellectual tasks they've been working on.
AI practitioners can apply their methods to virtually any area of human intellectual endeavour, making AI a universal field.
Artificial Intelligence seeks to enable machines to solve complex problems in ways reminiscent of human thinking.
AI empowers machines to perform tasks such as natural language understanding, pattern recognition, data-driven decision-making, and learning from experiences.
John McCarthy's definition of AI describes it as the science of creating intelligent machines, emphasizing both theory and practical implementation.
Russell and Norvig define AI as the study of agents that interact with their environment by receiving information and taking actions.
Elaine Rich's definition underscores AI's aim to make computers excel in areas where humans currently outperform, like understanding and producing language.
AI draws from various disciplines, including mathematics, software engineering, linguistics, and psychology, leading to diverse interpretations.
Roger C Schank identifies two main AI goals: creating intelligent machines and understanding the nature of intelligence.
Schank highlights communication, internal knowledge, world knowledge, intentionality, and creativity as key attributes of intelligence.
While AI has made progress in various areas, it still faces challenges, particularly in achieving human-level language skills and understanding of emotional context.
AI largely remains an evolving field, with definitions and goals subject to change as technology advances and AI reaches new stages of development.
References
Artificial Intelligence: A Modern Approach (1995) – Stuart J. Russel and Peter Norvig,
What is AI, Anyway? (1987) – Roger C. Schank
Artificial Intelligence (1983) – Elaine Rich
Introduction to Artificial Intelligence (2018) – Wolfgang Ertel