Identifying the Age of Incompetence:
David Cheriton argues we’re living in an ‘age of incompetence,’ where the increasing dependence on automation is highlighting the flaws of human performance.

He provides examples from various fields such as chess, Go, acrobatic maneuvers, and others where machines have outperformed humans.

Human Limitations:
Cheriton critiques humans as slow, lazy, error-prone, expensive, and sometimes dishonest.

An example from aviation highlights human error, as a pilot nearly caused a catastrophic disaster due to confusion and poor decision-making.

The Dominance of Operator Error:
Cheriton points out operator error as a dominant cause of issues in various fields, including commercial airline travel, train operations, and computer networks.

Even with systems that have fault tolerance, operator error remains a significant issue, especially in life-critical or business-critical areas.

The Decline of Human Competence:
Automation can often perform tasks significantly faster and with higher accuracy than humans, often with “five nines” or “six nines” of reliability.

Cheriton claims the cost of human labor is significantly higher due to its limitations and inefficiency.

He argues that human competence is declining over time. For instance, he notes that many people can’t survive without government assistance today, unlike previous generations.

Speed and Impact of Technological Advancements:
Cheriton discusses the increasing rate of automation and technology outpacing human ability in various fields. He acknowledges that technology is getting faster and better at most tasks, leading to a widening gap between human and machine abilities.

Technological Domination in Various Fields:
Cheriton highlights specific examples of areas where machines have surpassed humans, such as computation, chess, and airplane autopiloting.

He uses these as examples to illustrate how, over time, machines have gradually taken over tasks previously dominated by humans, often surpassing human performance levels.

Future Trajectories and the Singularity:
Cheriton briefly mentions the concept of the singularity as proposed by Ray Kurzweil, where human and machine intelligence merge.

However, he questions the relevance of human contribution in this scenario, given the speed and degree of technological advancements.

Reframing Technological Progress:
Cheriton suggests that the real measure of technological progress should not be how closely it can mimic human behavior (like flapping wings of a bird), but rather its utility and effectiveness in completing tasks (like the 747 airplane).

This reframing indicates that the focus should be on optimizing technology’s functionality over mirroring human attributes.

Automation and Human Intelligence:
Cheriton claims that human intelligence is not the goal, but rather the successful handling of tasks.

He proposes his “laws of automation” that state everything that can be automated will be automated and that everything can be automated.

This suggests a potential trend of increasing reliance on automation, resulting in humans becoming less proficient in certain tasks.

Software Smarter than Humans:
Cheriton counters the argument that humans must program computers, so they cannot truly take over. He argues humans can and have written software that outperforms them in specific tasks like chess and flying planes.

This phenomenon is linked to evolution, where less effective versions don’t make the cut, and the better ones survive and proliferate.

Test-Constrained Development and Software Evolution:
Cheriton asserts modern software development, such as test-driven development and continuous integration, is akin to evolution.

He likens developers making changes to software to mutations in evolution. These changes are vetted, and harmful “mutations” are weeded out, reflecting a sort of software evolution.

He labels this process “monkey-oriented programming,” where most changes (mutations) are harmful, and the survival challenge lies in weeding these out.

Limitations of Current Progress in Software:
Cheriton notes that despite the ongoing software evolution, there hasn’t been significant innovation in certain areas.

Existing operating systems, programming languages, and technologies (like Unix, C/C++, Java, SQL, TCP/IP, and X86 processors) have remained fundamentally the same for decades, with only cosmetic changes.

He suggests that this current state of affairs will not likely change in the near future.

Perceived Plateau in Programming:
David Cheriton suggests that software development may have plateaued, with fewer groundbreaking innovations in sight. He likens this to the evolution of computer chess, which has now reached a stage where only a handful of people can compete against advanced chess software.

Software Bugs and Performance Issues:
Cheriton states that despite technological advancements, software bug rates remain high, with no significant improvements over the years. He mentions buffer overflow as a persistent problem, unchanged since its appearance 40 years ago.

He also mentions performance issues in software and asserts that many modern software engineers are unaware of the costs of resources.

The Last Algorithm and the Rise of Machine Learning:
Cheriton introduces the idea of the ‘last algorithm’ as an indication of reaching the end of significant inventions in the software development domain. He suggests we may have reached a point of saturation for developing new algorithms and systems.

He highlights the rise of machine learning and unsupervised learning, which involves systems learning and improving without the need for direct human programming or intervention.

Automating Programming:
Cheriton discusses the possibility of automating programming, stating that some degree of this is already happening with machine learning and genetic programming.

He points out that current trends are leading to increased automation and possibly even to the ‘death of programming’.

Enduring Software:
According to Cheriton, modern software is often too costly and complex to replace, leading to software ‘living forever’. This means that instead of replacing old systems with new ones, we’re more likely to continue building upon and evolving existing systems.

Software Evolution:
Any software can theoretically be evolved, but in practice, some become too large and complex.

The architecture should be carefully maintained, ideally through modularity and good APIs.

Monkeys in Cages Analogy:
Cheriton suggests confining developers’ work to specific processes, similar to putting monkeys in separate cages.

Communication among processes should be kept simple and modular, similar to monkeys “screaming” at each other but not interacting physically.

This approach can be beneficial in distributed systems, and it helps manage complexity.

Self-Destructive Processes:
Processes should be engineered to “kill” themselves when they’re stuck or face a complex issue.

This technique simplifies problem solving and promotes the idea of process restart instead of complex debugging.

The effectiveness of this approach is highlighted through the example of ‘microrebooting,’ which can be more effective than making a system more solid or efficient through standard testing.

Rethinking Software Education:
Current software education often assumes students work on a blank slate and fully understand their software, which isn’t realistic in the real world.

It might be beneficial to teach students how to experiment with and evolve existing software, even if they don’t fully understand it.

This process-oriented approach can promote a more scientific mindset, focusing on controlled experimentation rather than complete understanding.

Importance of Testing:
Detecting when something breaks is crucial in this approach, emphasizing the importance of testing.

The concept of ‘live testing’ in a web context, merging development and operations, is a real-world application of this philosophy.

On API Development:
The importance of designing clear and well-structured APIs is often underscored, but this advice tends to be vague.

Cheriton appreciates CRUD (Create, Read, Update, Delete) as a principle of API design from the database world, which offers a practical guide to creating APIs.

State and Schema Challenges:
Managing state in applications is complex and challenging.

Relying on one schema throughout the development process is unrealistic because it’s hard to get it right.

Cheriton points to Kafka, an open-source system, as an effective approach. It allows developers to design their own schemas and databases based on a central logging service.

Software Janitor Specialty:
Cheriton introduces the concept of software janitor specialty, which is about cleaning up and maintaining existing software.

This task involves managing and reducing technical debt, which is a challenge due to the complexity of software systems.

Perception of Human Inadequacy:
– David Cheriton suggests that humans are, in many respects, already outperformed by computers. This isn’t limited to specialized fields, such as the game of Go, where computers have beaten world champions, but extends to broader areas like teaching computer science. Cheriton’s stance seems to imply that humans are less efficient, or “pathetic” as he puts it, compared to their computational counterparts.

Transition to a Predominantly Automated World:
– Cheriton underscores that the key shift occurs when computers outperform most people in certain tasks, not just the absolute best in the field. He suggests that in many areas, we have already reached this point. This transformation prompts us to reevaluate our roles and goals in a world where many tasks are better performed by computers.

Surviving in an Automated ‘Zoo’:
– Cheriton equates our current state to living in an “automated zoo”, encouraging us to embrace our new role and find satisfaction in it. While this could be seen as a demeaning perspective, Cheriton urges us to find contentment, even when the systems we create surpass us in various ways.

The Darwinian Evolution of Software:
– Cheriton likens the continuous advancement of technology to Darwinian evolution, emphasizing the survival of the fittest concept. He suggests our challenge is to navigate these changes effectively to prevent our obsolescence or “extinction.”

Recognizing and Preparing for the New World:
– Cheriton concludes by emphasizing the importance of acknowledging our current situation. While he doesn’t provide specific strategies, he implies that this acknowledgment is the first step in preparing for this “brave new world” where software outperforms humans in many domains.

Technology Evolution and Current Progress:
Cheriton acknowledges the advancements in technology, specifically in the areas of real-time speech recognition and image recognition, which have been facilitated by the development of new algorithms and technology improvements.

He points out that the availability of such advanced technology to ordinary people, such as facial recognition on phones, is impressive.

Impact of Automation on Society:
Cheriton argues that the development of more efficient tools through technology has made society more efficient, but it also highlights how much control we lack over our existence.

He notes that while society is designed by evolution rather than people, the risks associated with automation and the development of powerful technologies require careful consideration and respect.

Identifying Threats:
Cheriton advocates for an aggressive approach in identifying vulnerabilities in software, likening it to monkeys in a zoo trying to find weak points in their cages.

He cites the rise in cyber attacks as a wake-up call to address vulnerabilities in our systems.

Dependence on AI:
Cheriton recognizes the difficulties of understanding complex AI systems, like deep neural networks, which are not human-interpretable.

He asserts that while these systems may cause unforeseen problems, statistically they may still result in a net positive (such as reducing car crash fatalities).

The Future of Automation:
Cheriton suggests that while automation will never be perfect, it may still outperform humans in many areas.

He also posits a future where automation could help solve complex problems, improve living standards, and provide more leisure time, implying a potential for a high-quality human life in an automated future.