• Three remote lectures combining theory and practice
• Content will be released before each lecture
• You can ask questions at any moment
• Participation is key
• Contact by email: chc79@cam.ac.uk

An agent must find a way to reach a goal in its environment. But, the next step is not obvious.

A search problem is defined by a set of states, an initial state, a set of goal states, and a set of actions or transitions between states.

: set of states
: set of actions
: initial state
: set of goal states

A search problem is typically modelled using data structures such as graphs, adjacency matrices, queues and stacks.

"Within ten years a digital computer will be the world's chess champion." (Simon & Newell, 1958)

"Machines will be capable, within twenty years, of doing any work a man can do." (Simon, 1965)

"Within a generation... the problem of creating 'artificial intelligence' will substantially be solved." (Minsky, 1967)

"In from three to eight years we will have a machine with the general intelligence of an average human being." (Minsky, 1970)

"Within ten years a digital computer will be the world's chess champion." (Simon & Newell, 1958)

"Machines will be capable, within twenty years, of doing any work a man can do." (Simon, 1965)

"Within a generation... the problem of creating 'artificial intelligence' will substantially be solved." (Minsky, 1967)

"In from three to eight years we will have a machine with the general intelligence of an average human being." (Minsky, 1970)

"Within ten years a digital computer will be the world's chess champion." (Simon & Newell, 1958)

"Machines will be capable, within twenty years, of doing any work a man can do." (Simon, 1965)

"Within a generation... the problem of creating 'artificial intelligence' will substantially be solved." (Minsky, 1967)

"In from three to eight years we will have a machine with the general intelligence of an average human being." (Minsky, 1970)

"Within ten years a digital computer will be the world's chess champion." (Simon & Newell, 1958)

"Machines will be capable, within twenty years, of doing any work a man can do." (Simon, 1965)

"Within a generation... the problem of creating 'artificial intelligence' will substantially be solved." (Minsky, 1967)

"In from three to eight years we will have a machine with the general intelligence of an average human being." (Minsky, 1970)

Computing power, algorithms, and data were insufficient to solve real-world problems.

Combinatorial explosion

Easy tasks for humans are difficult for AI (Moravec's Paradox)

Computing power, algorithms, and data were insufficient to solve real-world problems.

Combinatorial explosion

Easy tasks for humans are difficult for AI (Moravec's Paradox)

Computing power, algorithms, and data were insufficient to solve real-world problems.

Combinatorial explosion

Easy tasks for humans are difficult for AI (Moravec's Paradox)

Computing power, algorithms, and data were insufficient to solve real-world problems.

Combinatorial explosion

Easy tasks for humans are difficult for AI (Moravec's Paradox)

Agents are limited because of the problem's complexity. They should leverage human knowledge and emulate human reasoning.

An expert system is defined by a knowledge base, inference engine, and user interface that work together to apply domain expertise to specific problems.

: Knowledge Base
: Inference Engine
: User Interface

Knowledge is typically modelled using subject, object, predicate semantic triple model.

Rules are if statements.

DENDRAL (1960s)

Identified unknown organic molecules using knowledge of chemistry

MYCIN (Early 1970s)

Supported bacterial infections diagnosis and treatment

XCON/R1 (1982)

eXpert CONfigurer - Automated the configuration of VAX computer systems (successful deployment)

PUFF (1982)

Interpreted pulmonary function test results to diagnose lung disorders

PROSPECTOR (1986)

An expert system for mineral exploration

DEEP BLUE (1997)

An expert system that defeated a chess world champion

"In medicine, management, and the military — indeed in most of the world's work — the daily tasks are those requiring symbolic reasoning with detailed professional knowledge." (Feigenbaum, 1982)

"Commercialising Artificial Intelligence." (The New York Times, 1982)

"Gains are Slow for Artificial Intelligence Industry." (The New York Times, 1987)

"New expert systems companies were being formed at a rate of what seemed like one a week. " (Hart, 2021)

"In medicine, management, and the military — indeed in most of the world's work — the daily tasks are those requiring symbolic reasoning with detailed professional knowledge." (Feigenbaum, 1982)

"Commercialising Artificial Intelligence." (The New York Times, 1982)

"Gains are Slow for Artificial Intelligence Industry." (The New York Times, 1987)

"New expert systems companies were being formed at a rate of what seemed like one a week. " (Hart, 2021)

"In medicine, management, and the military — indeed in most of the world's work — the daily tasks are those requiring symbolic reasoning with detailed professional knowledge." (Feigenbaum, 1982)

"Commercialising Artificial Intelligence." (The New York Times, 1982)

"Gains are Slow for Artificial Intelligence Industry." (The New York Times, 1987)

"New expert systems companies were being formed at a rate of what seemed like one a week. " (Hart, 2021)

"In medicine, management, and the military — indeed in most of the world's work — the daily tasks are those requiring symbolic reasoning with detailed professional knowledge." (Feigenbaum, 1982)

"Commercialising Artificial Intelligence." (The New York Times, 1982)

"Gains are Slow for Artificial Intelligence Industry." (The New York Times, 1987)

"New expert systems companies were being formed at a rate of what seemed like one a week. " (Hart, 2021)

"In medicine, management, and the military — indeed in most of the world's work — the daily tasks are those requiring symbolic reasoning with detailed professional knowledge." (Feigenbaum, 1982)

"Commercialising Artificial Intelligence." (The New York Times, 1982)

"Gains are Slow for Artificial Intelligence Industry." (The New York Times, 1987)

"New expert systems companies were being formed at a rate of what seemed like one a week. " (Hart, 2021)

Extracting knowledge from human experts and encoding it into rules was difficult, time-consuming, and expensive

Systems could not reason beyond their pre-programmed knowledge and failed when confronted with unexpected situations

Adding more rules often led to rule interaction problems and combinatorial explosion

Updating knowledge bases as domains evolved required significant effort

Agents cannot be fully pre-programmed. Agent must learn from examples to perform a task.

Agents cannot be fully pre-programmed. Agent must learn from examples to perform a task.

Learning from examples requires a function that approximates patterns in data, using a learning algorithm, and evaluation criteria.

: dataset of examples or pairs
: hypothesis space of possible functions
: loss/reward function measuring success
: algorithm to search through

Agents cannot be fully pre-programmed. Agent must learn from examples to perform a task.

Learning from examples requires a function that approximates patterns in data, using a learning algorithm, and evaluation criteria.

: dataset of examples or pairs
: hypothesis space of possible functions
: loss/reward function measuring success
: algorithm to search through

Learning algorithms use different computational representations during the learning and inference: dataframes, tuples, trees, graphs, matrices, etc.

Large and complex datasets that improve models' statistical power

• Social Networks
• Mobile Computing
• Internet of Things
• ...

Everyday Applications

• Voice and Chat Assistants (Siri, Chat-GPT)
• Content Recommendations (Netflix, Spotify)
• Smart Home Devices
• Facial Recognition
• Language Translation
• Fraud Detection
• Personalised Marketing
• Gaming
• ...

Promises

• Early diagnosis and personalised medicine
• Drug discovery
• Climate change solutions
• Personalised learning
• Accelerating science
• Productivity boost, automation
• Smart homes, cities, etc.
• ...

Risks

• Bias and discrimination, privacy violations
• Job displacement, digital divide
• Data quality issues, system vulnerabilities
• Adversarial attacks, data breaches
• Implementation costs, market disruption
• Energy consumption, carbon footprint
• Regulatory compliance, liability issues
• ...

Risks

• Bias and discrimination, privacy violations
• Job displacement, digital divide
• Data quality issues, system vulnerabilities
• Adversarial attacks, data breaches
• Implementation costs, market disruption
• Energy consumption, carbon footprint
• Regulatory compliance, liability issues
• ...

"We should be able to do 90 percent of miles driven [autonomously] within three years." (Musk, 2013)

"From a technology standpoint, Tesla will have a car that can do full autonomy in about three years, maybe a bit sooner." (Musk, 2015)

"I feel very confident predicting that there will be autonomous robotaxis from Tesla next year." (Musk, 2019)

...

"We should be able to do 90 percent of miles driven [autonomously] within three years." (Musk, 2013)

"From a technology standpoint, Tesla will have a car that can do full autonomy in about three years, maybe a bit sooner." (Musk, 2015)

"I feel very confident predicting that there will be autonomous robotaxis from Tesla next year." (Musk, 2019)

...

"We should be able to do 90 percent of miles driven [autonomously] within three years." (Musk, 2013)

"From a technology standpoint, Tesla will have a car that can do full autonomy in about three years, maybe a bit sooner." (Musk, 2015)

"I feel very confident predicting that there will be autonomous robotaxis from Tesla next year." (Musk, 2019)

...

"I predict that there will be millions of Teslas operating fully autonomously in the second half of next year." (Musk, 2025)

"We should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better than radiologists." (Hinton, 2016)

"We should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better than radiologists." (Hinton, 2016)

"AI will be capable of generating novel insights next year." (Altman, 2025)

"Self-improving AI will create a super intelligence." (Musk, 2025)

"When I look at the data, I see many trend lines up to 2027." (Clark, 2025)

"There is a 10-20% chance that the technology will end in human extinction." (Hinton, 2025)

"Our relationship with future AI systems is that we are going to be their boss." (LeCun, 2025)

"Machines will be capable, within twenty years, of doing any work a man can do." (Simon, 1965)

"In from three to eight years we will have a machine with the general intelligence of an average human being." (Minsky, 1970)

"In medicine, management, and the military — indeed in most of the world's work — the daily tasks are those requiring symbolic reasoning with detailed professional knowledge." (Feigenbaum, 1982)

"AI will be capable of generating novel insights next year." (Altman, 2025)

"Self-improving AI will create a super intelligence." (Musk, 2025)

"When I look at the data, I see many trend lines up to 2027." (Clark, 2025)

"There is a 10-20% chance that the technology will end in human extinction." (Hinton, 2025)

"Our relationship with future AI systems is that we are going to be their boss." (LeCun, 2025)

"AI will be capable of generating novel insights next year." (Altman, 2025)

"Self-improving AI will create a super intelligence." (Musk, 2025)

"When I look at the data, I see many trend lines up to 2027." (Clark, 2025)

"There is a 10-20% chance that the technology will end in human extinction." (Hinton, 2025)

"Our relationship with future AI systems is that we are going to be their boss." (LeCun, 2025)

"... step 1: solving intelligence, step 2: use it to solve everything else..." (Hassabis, 2025)

"Artificial General Intelligence (AGI) will emerge in the next five or 10 years." (Hassabis, 2025)

Machine Learning Problem Requirements

1. Massive combinatorial search space

2. Clear objective function (metric) to optimise against

3. Lots of data and/or accurate and efficient simulators

Machine Learning Problem Requirements

1. Massive combinatorial search space

2. Clear objective function (metric) to optimise against

3. Lots of data and/or accurate and efficient simulators

Machine Learning Problem Requirements

1. Massive combinatorial search space

2. Clear objective function (metric) to optimise against

3. Lots of data and/or accurate and efficient simulators

Machine Learning Problem Requirements

1. Massive combinatorial search space

2. Clear objective function (metric) to optimise against

3. Lots of data and/or accurate and efficient simulators

"It seems to me what is called for is an exquisite balance between two conflicting needs: the most skeptical scrutiny of all hypotheses that are served up to us and at the same time a great openness to new ideas. Obviously those two modes of thought are in some tension. But if you are able to exercise only one of these modes, whichever one it is, you’re in deep trouble. (The Burden of Skepticism, Sagan, 1987)

Learning Model	Accuracy
Model v1	88%
Model v2	90%
Model v3	98%

Learning Model	Accuracy
Model v1	88%
Model v2	90%
Model v3	98%

Learning Model	Accuracy
Model v1	88%
Model v2	90%
Model v3	98%

Does Model v3 satisfy users' requirements?

Learning Model	Accuracy
Model v1	88%
Model v2	90%
Model v3	98%

Does Model v3 satisfy users' requirements?

For example:

• Users require accurate and fast responses
• The project budget is limited
• The system must be user-friendly
• Users with disabilities must be prioritised
• ...

Threat to AI applications and promises

• Mundane applications for problems we did not know we had
• Disregard for social and environmental implications
• Unrealistic expectations and hype
• Exclusion of diverse perspectives and voices
• Unsustainable technologies
• Increased inequality and digital divide
• Security and privacy concerns
• ...

"Grok referenced the debunked conspiracy theory of “white genocide” in South Africa in reply to unrelated prompts, such as questions about baseball or humorous requests to talk like a pirate. " (May, 2025)

"ChatGPT suggested a dangerous route to tourists hiking in the mountains in Poland and resulted in the hikers having to be rescued." (January, 2025)

"Steve Talley, a financial adviser in Denver, Colorado, was wrongfully arrested twice for bank robberies he did not commit, based on flawed facial recognition technology and questionable identification procedures." (September, 2014)

Learn more at the AIAAIC Repository

Latency Requirement

(< 5 secs)

Learning Model	Accuracy
Model v1	88%
Model v2	90%
Model v3	98%

Latency Requirement

(< 5 secs)

Learning Model	Accuracy	Latency
Model v1	88%	3 secs
Model v2	90%	4 secs
Model v3	98%	10 secs

Resource Limitations

(Low budget)

Learning Model	Accuracy	Latency
Model v1	88%	3 secs
Model v2	90%	4 secs
Model v3	98%	10 secs

Resource Limitations

(Low budget)

Learning Model	Accuracy	Latency	Resource Demand
Model v1	88%	3 secs	Low
Model v2	90%	4 secs	Medium
Model v3	98%	10 secs	High

Resource Limitations

(Low budget)

Learning Model	Accuracy	Latency	Resource Demand
Model v1	88%	3 secs	Low
Model v2	90%	4 secs	Medium
Model v3	98%	10 secs	High

The Ecosystem:

• Community of universities around the continent
• Inclusive training programmes
• Technical knowledge
• Students with initiative

Social Problems:

• Food security and hunger
• Poverty and inequality
• Healthcare issues (e.g., epidemics)
• Access to clean water and sanitation
• Gender inequality and violence
• Access to quality education
• Climate change and environmental problems
• ...

This execution time does not suit low-latency requirements, but that is how ACO is designed.

This execution time does not suit low-latency requirements, but that is how ACO is designed.

Selecting edge servers close to current and future users' location. We used two approaches that cluster historical trips and use these clusters to predict the next link in the user's path:

• Bayesian Classifier
• Hidden Markov Model

Overview

• ML Context
• ML Today
• ML Perception
• ML Objective Definition
• Context Matters
• DSA Example
• Sercice Placement Example

Overview

• ML Context
• ML Today
• ML Perception
• ML Objective Definition
• Context Matters
• DSA Example
• Sercice Placement Example

Next Time

• ML Adoption Process
• Problem First
• The Data Science Process
• Data Orientation
• Data Quality
• A Machine Learning Pipeline