Into the world of modern AI

In this section, we will explore the history of AI in three sections. We will begin by looking at the initial applications of AI systems. Then, we will move to a critical (often overlooked) period of AI that we have been living with since the early 2010s. This will be followed by ‘modern’ AI, which we will assume to have begun in 2017 with the advent of the Transformer.

So, before we take a deeper look into how we got to where we are now, it would be useful to take stock of what we have learned so far about AI. An excellent and low-intensity introduction to AI as a technology is provided in the Philosophize This! podcast, Episode 185, entitled “Should we prepare for an AI revolution?“. 

TASK: Consider the following tasks as we work through this next section of the course.

1. Industrial Revolution Redux
List two ways today’s AI expansion resembles the first Industrial Revolution and two ways that it differs.

2. From Ranking to Generating
Ranking algorithms reorder what humans create; generative models create new content outright. Which stage has had (or will have) the larger impact on everyday life, and why?

3. The Responsibility‑Race Dilemma
In Aza Raskin and Tristan Harris’s viral presentation for the Center for Humane Technology, the pair claim that new technologies (a) create fresh responsibilities, (b) trigger a competitive race for power, and (c) risk tragedy without coordination.
Choose one AI sphere such as social‑media feeds, chatbots, medical applications, education and briefly specify:
• one new responsibility (and opportunity) it imposes
• one way the competitive race is visible
• one coordination step that could avert harm

4. Witnessing Your Own Obsolescence
If a skill you spent years mastering were to be automated tomorrow, how would it affect your sense of identity and purpose? 

A Brief History of AI

Artificial Intelligence has come a long way since its inception, evolving from simple rule-based systems to complex neural networks capable of human-like (or imitating human-like) reasoning. The history of AI dates back to the 1950s when the term was first coined and was followed by early AI systems which achieved some success across numerous applications but faced challenges due to limited computational power and data. 

The following excellent history of AI, “Human Language Understanding and Reasoning” by Christopher Manning, published in 2022 in Daedalus, provides useful background to these AI systems.

So, what has allowed for the seemingly recent explosion of AI?

Image from BHC Library

Initial Applications of AI

Initially, AI excelled in tasks with clearly defined rules and structures. An illustrative example is IBM’s Deep Blue, which defeated chess world champion Garry Kasparov in 1997, demonstrating AI’s potential in solving complex problems with well-articulated rules. 

 Image from Stan Honda

For example, looking at the chess rating achievable by a computer, the increase has been relatively consistent over the past decades. The best computer now is almost 30% better than Deep Blue, which was released almost three decades ago. That is a 10% improvement every 10 years.

AI in the 2010s – Big Data and the Ranking Algorithm as a Case Study

In this section, we will look at a select example of how AI has been used since the 2010s. This was a time characterised by big data, but before the modern AI algorithms we are working with today. As you work through this short section, you are encouraged to think about how long ago the early 2010s were and how long we have had to gather evidence on how the introduction of AI into our daily lives has changed our society.

By the early 2010s, artificial intelligence had transitioned from specialised applications to becoming an integral part of everyday digital interactions. This shift was facilitated by several converging factors: the widespread adoption of smartphones, the exponential growth of user-generated data, and advancements in machine learning algorithms capable of processing these data at scale.

Social media platforms, notably Facebook (now Meta), began employing sophisticated AI-driven ranking algorithms to personalise user content feeds. The initial algorithm, known as EdgeRank, considered factors like user affinity, content weight, and time decay to determine the visibility of posts in a user’s feed. As data volumes grew, these hand‑tuned rules were replaced by machine‑learning models that learn, from billions of clicks, which items will keep each user scrolling. This progression of AI through our social spheres is brilliantly captured by the Center for Humane Technology’s award-winning documentary film, “The Social Dilemma“.

Because engagement became the yardstick of success, AI models steadily learned to surface content that provokes strong reactions such as surprise, anger, or envy, because these emotions are reliably ‘rewarded’ with likes, shares, and comments. 

Case Study 1: ‘The Great Rewiring’

Jonathan Haidt, in his 2024 New York Times Bestseller, “The Anxious Generation” labels this migration into the digital world, “The Great Rewiring”, evidencing that AI-curated feeds have had a measurable impact on adolescent well‑being. 

Case Study 2: ‘The Great Hack’

The Cambridge Analytica scandal illustrates how the same ranking infrastructures used for social media algorithms, coupled with detailed behavioural data, can be used to target political messages at scale.

TASK: The two previous case studies highlight some of the more insidious applications of AI from the last 20 years. 

Research how governments, collectives and corporations have reacted to this evidence to produce positive change.

Applying AI in Different Domains and the Transformer

A real challenge is expanding AI’s capabilities to domains involving ambiguity and tasks lacking explicit rule-based structures. This may involve tasks such as image, speech & handwriting recognition, reading comprehension & language understanding, and predictive reasoning. 

Take a look at the below figure, published in 2024 by Our World in Data, that examines how AI systems have scored over time, compared to human performance on various tasks. Note that the human baseline is set to ‘0’ and initial AI capabilities are set to ‘-100’. 

Image from Our World in Data

The figure shows the extremely rapid advances of the abilities of AI in fields that lack well-articulated rules. It is also worth noting that amongst the collection of tasks without well-defined rules, advancements appear to be moving in unison.

Imagine trying to write a rule-book for speech recognition or complex reasoning. Now imagine trying to turn that into instructions you could give to a computer (1’s and 0’s). Is this what happens in these AI systems?

TASK: Use the following prompt in an AI system of your choosing, “In under 50 words, explain how AI manages tasks that lack clear rules, such as image or speech recognition”.

In the next section, we will look at the four key ingredients that came together in 2017 to allow for a huge explosion in AI.