The Evolution of Artificial Intelligence

The recent excitement surrounding ChatGPT and OpenAI has catapulted Artificial Intelligence (AI) into the public spotlight (see Figure 1). Within this update, we will discuss the evolution of AI and its core concepts and will follow up next month with a broad discussion of Fairlight’s exposure to the technology.

Figure 1.

History of AI

In 1950, Alan Turing published a paper discussing seminal Computing Machinery and Intelligence, where he opened with the sentence of “I propose to consider the question, can machines think?” He developed a test called “The Imitation Game”, also known as “The Turing Test”. The game was relatively simple: a human participant (the interrogator) would exchange a series of typed interactions with two parties behind a partition. One of these parties was a human, and the other was a machine. The goal of the game was for the interrogator to identify which party was human. If the interrogator failed to do so, the machine would be considered to have been thinking, and the winner. This event is widely regarded as the birth of AI.

Over the next 70 years, we witnessed rapid development in AI and machine learning. This included the ELIZA chatbot in the 1960s, the introduction of backpropagation algorithms in the 1970s, the AI winter of the 1980s, IBM’s Deep Blue defeating world chess champion Garry Kasparov in 1997, AlphaGo defeating legendary Go player Lee Sedol in 2016, and most recently, OpenAI’s ChatGPT in 2022.

The world of AI is vast, with formal branches difficult to define given many of the underlying technologies have cross over techniques integrated from various verticals. However, we can identify core concepts and their use cases to better understand the evolving landscape.

AI: Machine Learning, Natural Language Processing, Computer Vision, Expert Systems, and Robotics

Machine Learning is considered one of the biggest branches of AI. It is a subset of AI that enables computers “to learn” from data and make decisions based on that data. Many techniques from machine learning are used in other technologies. Machine learning can be categorized into three types:

• Supervised learning relies on labelled data. Based on the input and output given by humans, the computer will derive the model. An example of this is the classification of email as spam or not spam.

• Unsupervised learning uses unlabelled data and relies on algorithms to find patterns and relationships. In this case, humans will only give the input as the data source. An example of this is anomaly detection in banking, where the algorithm will identify unusual spending patterns that could indicate fraudulent transactions.

• Reinforcement learning involves training the models to make decisions by rewarding or punishing the model to achieve an optimal result. It mimics the human process of trial-and-error. An example of this is the advancing development of autonomous driving.

Natural Language Processing (NLP) is a branch of AI that focuses on the interaction between computers and humans through the exchange of natural language. The goal of NLP is to enable computers to understand, interpret, and interact with human language in a way that is understandable and meaningful. Examples of NLP applications include ChatGPT, which can engage in conversations with its users. Another example is speech recognition systems like those used in virtual assistants.

Computer vision involves enabling computers to interpret and make decisions based on the visual data. This branch allows the machine to recognize objects, understand scenes, and extract information from images and videos. One very common example is facial recognition in mobile phones.

Expert systems are AI programs that mimic the decision-making abilities of human experts. They use a knowledge base of human expertise and using an inference engine, they solve specific problems within the expert domain. Expert systems are generally rule-based “if-then statements”. For example, a medical diagnosis will use fact databases and then use if-then logic to infer diseases from the symptoms.

Robotics, as the name implies, is a branch of AI that deals with the design and creation of robots. AI-powered robotics have significantly improved efficiency in many industries by automating complex, tasks, reducing human error, and improving safety.

The Fairlight View

Whilst AI has attracted more attention recently, this isn’t the first time that our portfolio companies have had to face technological change. In fact, AI is not particularly new for Fairlight’s portfolio companies. Quality businesses tend to find ways to adopt new technologies to improve their competitiveness rather than dismissing them. Next month, we will discuss how some of our investment companies have been and will likely be impacted by the improvement of AI, and why our preference is to sell the “picks and shovels” in the AI gold rush.

‍

‍