Technical Breakdown: Lighthill Report
The First AI Winter: How Hype and Unrealistic Promises Led to a Collapse in Funding
Summary
Lighthill Report: AI research split into three categories—Automation, CNS Research, and the Bridge.
Limited Progress: AI struggled with general problem-solving and relied heavily on human input.
Overhyped Claims: Predictions of AGI by 1980 and ASI by 2000 were unrealistic.
First AI Winter: Funding cuts in the UK and U.S. led to stalled AI progress.
Future Uncertain: Another AI winter is possible, but rapid advancements suggest otherwise.
First AI Winter
The field of AI was advancing rapidly, but was this due to increased resources or genuine breakthroughs in the field?
In 1972, the Science Research Council (SRC) was receiving a barrage of applications for grants and resources from researchers in the field of AI. To determine if this increase in popularity in the field was warranted, they asked James Lighthill, a respected mathematician, to review the state and give a general view of the field.
Lighthill was tasked with studying the literature in the field for two months and speaking with a variety of AI researchers before writing this report.
He spanned all the research done in the field since its inception (1950-1972) to make this report.
Before he began writing, he stated:
“The report which follows must certainly not be viewed as more than such a highly personal view of the AI field”.
Three Segments of AI
Lighthill decidedly separated AI into three different sections he called ABC:
Advanced Automation
Bridge Between
Computer Based Central Nervous System (CNS) Research
Advanced Automation
The objective of this category is to replace human beings with machines for specific tasks. The use cases span a variety of areas such as:
Industrial: machine recognition of text and voice, and design and assembly of whole projects.
Military: Cryptography and guided missiles.
Mathematics: Automation of problems of logical deduction including theorem proving.
The two dominant themes in this category are information retrieval and problem solving.
Information retrieval depends on a “knowledge base” (data) and the “file structure” of this knowledge base.
Problem-solving goes beyond math, addressing real-world challenges like decision-making and interpreting ambiguity.
In the long term, Lighthill envisioned these two dominant themes merging, leading to the development of programs with enhanced learning capabilities.
Computer Based CNS Research
This component is concerned with the theoretical investigations related to neurobiology and psychology.
It was specifically used to interpret large amounts of neurobiological data in specific areas of the CNS using neural networks. This was done to understand how we function through computer simulations.
As opposed to mimicking human behaviour, this category is about understanding emotional or perceptual data gathered from psychologists and neurobiologists.
Bridge
The bridge is concerned with combining both of the previous sections to create an automatic device that mimics a certain range of human functions: A Robot.
He argues that the justification of the resources being given to the research done in this field depend on the fundamental coherence of the whole field of AI, spanning all three categories. More specifically, the bridge between must be viewed as important enough to be justified as a link between A and C.
Lighthill’s view
Lighthill then compared the achievements in each category to conventional advancements made over the same time frame—the past 25 years since the start of AI research. These conventional achievements included innovations developed using traditional computing methods without AI or those derived through manual calculations.
Advanced Automation
Lighthill concluded that category A techniques are not successful when they are developed with a high degree of general applications. They have only proven useful when given a large quantity of knowledge about a specific problem with narrow breadth.
He notes that AI advancements in pattern recognition, speech recognition, machine translation, and mathematical discovery have been underwhelming. While they have innovated the field in some ways, they haven’t delivered anywhere near what the promises held 25 years prior.
Computer Based CNS Research
While there have been some discoveries in this category, the problems with them arise due to their theoretical nature. Neural networks were used to uncover theoretical relations in things like associative recall, classification, and inductive generalization.
The problem is that the biggest discoveries in the field did not come from computer modelling at all, making these advancements pale in comparison to conventional ways.
He also said that researcher’s hopes of having these models go beyond memorizing facts to being more general was a dream that was untimely and unpromising:
“One can only conclude the nineteen-seventies are not the right decade in which to begin researches aimed at applying such techniques…”
Bridge
The bridge was where Lighthill said lived the biggest disappointments.
Lighthill discussed AI robots playing chess, noting that they performed at the level of an experienced amateur at the time. However, compared to conventional computer programs, they were slower, less effective, and less reliable.
He also noted some of the promises made like “possibilities in the nineteen-eighties include an all-purpose intelligence on a human-scale knowledge base;... possibilities suggest themselves based on machine intelligence exceeding human intelligence by the year 2000…” These claims of Artificial General Intelligence (AGI) by 1980 and Artificial Super Intelligence (ASI) by 2000 were outlandish.
The high expectations, contrasted with the limited progress in areas like hand-eye coordination and common-sense problem-solving, led to widespread disappointment.
This led Lighthill to argue that AI’s bridge was failing to connect automation and cognitive research.
First AI Winter
AI winter is a term used to describe a decrease in funding and hype surrounding AI. The result of this paper led to the first AI winter, as the UK and the U.S. reduced their funding for AI research due to frustrations over the lack of advancements.
James Lighthill’s paper was critical in causing this, as his report was what led the UK government to reduce their funding in the first place.
This resulted in many AI projects at the time getting slowed down or canceled altogether as researchers struggled to secure grants.
With massive investments in AI, some fear another AI winter may be inevitable. If the advancements fail to deliver on their promises, we might repeat history, causing a potential pause in AI research.
However, with the rapid advancements of AI nowadays, this potential winter does not look to be coming soon. New advancements seem to be happening every week (Check out the AI Weekly Briefing for up-to-date news) and may never stop coming.
Does this mean we’re in the endgame of AI advancements?