Introduction of Danny Hillis:
Danny Hillis, a well-known figure in the tech world, trained under Claude Shannon and Marvin Minsky, pioneers in the field of computing. Hillis himself made substantial contributions in the development of key concepts for supercomputing and data storage techniques, and currently works at Applied Minds, a company focusing on creating new inventions.

The Evolution of Artificial Intelligence (AI):
AI’s evolution can be segmented into different areas: cybernetics, classical AI, neural networks, and big data.

AI has been redefining what we consider “thinking”. Initially, any machine behavior that exhibited any form of intelligence was credited as thinking. Over time, as machines evolved to handle more complex tasks, our definition of thinking shifted.

Shifts in AI Development:
Early AI development focused on tasks such as auto-piloting and error correction, which were initially thought to require thinking. Once machines were able to perform these tasks, they were no longer considered acts of thinking.

The next shift in AI involved symbol processing or the understanding of human symbols, a feature now commonly used in Google searches.

AI’s next stage involved tasks that previously were considered only capable of human intellect, like playing chess.

Understanding Human and Machine Thinking:
Human thinking is just one form of thinking, and AI has revealed the usefulness of having machines that think differently than humans, providing a complementary perspective.

Initial AI research focused on tasks humans found difficult like playing chess or solving calculus problems. More recent research has shifted towards easier tasks that involve more “common sense” reasoning, such as face recognition or understanding basic principles like a bottle spilling when tipped over.

The Fragmentation of AI:
The concept of a singular, all-capable AI has fragmented into multiple AI entities, each specializing in different tasks.

Much of AI research has been about understanding the nature of thinking itself. It initially focused on the challenging aspects of human cognition, but over time, it became apparent that the more “automatic” parts of thinking, like basic reasoning and language understanding, are critical and complex processes to replicate in AI.

AI’s Role in Game Domains:
AI has greatly advanced into various fields, one of them being games like chess. With computers now capable of beating grandmasters, the presence of AI didn’t eliminate chess as a human game, indicating an adaptability of humans to AI mastery.

Shift in Perception Towards AI:
There has been a significant shift in public perception of AI. In the past, AI was seen as threatening, potentially replacing jobs or even creating destructive super-beings. However, now, as seen in events like Watson’s victory in Jeopardy, people are increasingly rooting for the machines.

This shift is largely due to widespread use of intelligent machines in everyday life, such as cellphones and personal computers. People have begun to see these devices as life-enhancing rather than threatening.

AI as an Extension of Human Abilities:
As machines have become more intelligent, being compared to a machine is less insulting and more often viewed as a compliment. Machines are seen as extensions of human capabilities, complementing and enhancing our abilities.

The Question of Replacement and Autonomy:
The gradual integration of AI in tasks, such as driving, raises the question of whether people want to be aware when they’re being replaced or helped by AI. The balance between control and convenience varies among individuals, with some preferring direct control (e.g., manual car operation) and others welcoming the assistance (e.g., self-driving cars).

However, it’s observed that for the current generation, in most circumstances, convenience seems to be favored over complete control.

Motivation to Build Thinking Machines:
Hillis expressed a longstanding desire to build a computer that could feel pride, a concept indicative of a machine that possesses deep learning and thinking capabilities.

This idea aligns with an age-old ambition of creating entities with capabilities similar to human cognition, providing a sense of intellectual accomplishment and progeny.

Ultimate Complexities: Life and Thinking:
Two of the most complex phenomena in the world are life and thinking. These complexities attract individuals with an engineering mindset, as understanding and potentially creating these phenomena represent the ‘holy grails’ of engineering and invention.

Creating life and thinking machines are both at their infancy stages, marking the beginning of an exciting journey for Hillis and others in his field.

Spiritual Aspect of Creating Thinking Machines:
Hillis acknowledges a spiritual aspect in his desire to build thinking machines. While most technological development is justified by its utility or potential to improve lives, building thinking machines carries a deeper philosophical and spiritual appeal.

Even though they also believe or convince themselves that such advancements will contribute to a better world, there’s an undeniable allure and excitement in the adventure of grappling with ultimate complexities.

Technological Progress in Artificial Intelligence:
A lot of progress in self-driving cars and AI in general has been due to the increase in computation speed. This has allowed us to implement ideas from the 60s and 70s in real-time, such as processing 60 frames per second from multiple cameras and laser rangefinders on vehicles.

The Impact of DARPA Challenge:
The DARPA Challenge of 2004 focused attention on the problem of creating self-driving cars. Participants found ways to use a laser rangefinder to get 3D information directly, which gave a significant boost to the development of self-driving cars.

The Evolution of Driving:
As technology progresses, traditional controls like the brake and steering wheel will be used less and less. Instead, these commands will be processed by an onboard computer that calculates the best actions. Eventually, you will simply tell the car where you want to end up, and the computer will handle the rest.

The Role of High-Resolution Maps:
High-resolution maps, such as those created by Google, are currently used by self-driving cars to make it easier for the computer to identify key objects, like street lights, and to navigate. However, this is not a necessity, and future self-driving cars might not need these maps, relying instead on their onboard sensors and processing capabilities.

Safety and Efficiency Improvements:
Hillis argues that self-driving cars will be significantly safer than human-driven ones, potentially reducing the tens of thousands of annual traffic deaths. Additionally, self-driving cars will be able to utilize road space more efficiently, potentially eliminating traffic jams and the need for parking.

Gradual Transition towards Self-Driving Cars:
The transition to self-driving cars will not be sudden. People will gradually get used to being driven by computers, as more and more traditional car controls are replaced by automated systems.

Dual Nature of Intelligence:
Hillis highlights the existence of two schools of thought in the field of Artificial Intelligence: symbol processors, who write intricate programs, and neural network enthusiasts, who train machines to learn through examples without fully understanding how this learning occurs.

After much debate, Hillis suggests the consensus is that intelligence involves both symbolic and neural network-like processing, reflecting Daniel Kahneman’s concept of “thinking fast and slow”.

Subconscious vs. Conscious Thinking:
Most of our thinking, according to Hillis, happens subconsciously, using heuristics or “trained kind of example thinking,” and isn’t based on logic.

Another part of our thinking is logical, like a chess player, involving planning, making deductions, and imagining a perfect world. This kind of thinking is just the “tip of the iceberg.”

Hillis suggests the “logical” thinking, the voice in our head we often equate with intelligence, isn’t the majority of our thinking process, and it’s possible to function without it.

Implications for AI Development:
Hillis provides the example of Watson, an AI whose architecture embodies both types of thinking. Watson comprises numerous weak programs that didn’t work well individually, but an executive function on top of them was able to weigh the inputs from these programs and generate an answer.

Perception as Controlled Hallucination:
The development of AI has influenced cognitive science, leading to fascinating insights about human perception. Our brains use predictions to help our eyes know what to expect when processing visual stimuli.

This interplay between perception and prediction suggests our vision is a “controlled hallucination,” where we’re projecting our mental image onto the world and adjusting it when reality doesn’t match.

The idea of our perception being a hallucination extends to other levels, indicating our perception of reality, although concrete to us, is essentially a construct of our mind.

Color Perception as a Construct:
Hillis highlights the subjective nature of even basic elements of perception, such as color, by referencing cultures without a word for the color blue. People from these cultures often fail to distinguish blue from green, indicating that even color perception is a kind of “hallucination”.

Proteomics and AI for Predictive Healthcare:
Hillis emphasizes the need for more predictive healthcare, using his work with proteomics and AI as a potential pathway to that goal. Proteomics studies proteins, the body’s signaling system, which can reflect the body’s condition and potential medical issues. His team has developed a way to extract a significant amount of data from a simple drop of blood, allowing them to detect health issues such as colon cancer before they become apparent. This approach could shift healthcare from treating illnesses to preventing them in the first place.

The Future of Healthcare:
Hillis imagines a future where individuals regularly take their own health samples, similar to diabetics checking their blood sugar levels. They would then send these samples to be analyzed for early signs of health issues. Through small and large interventions based on these early signs, individuals could prevent or mitigate health issues before they become severe.

Innovation Path and Process:
Hillis explains that innovation follows a process with several stages: conceptualization, initial trials, working prototypes, commercial successes, and widespread adoption. Using the light bulb as an example, Hillis illustrates that countless failed attempts often precede a successful innovation. He states that the field of proteomics is currently in the “hundreds of light bulbs” stage, where many different ideas and methods are being explored.

Surprises in Technology Development:
Hillis discusses that what often surprises him is not the initial idea, but rather the successful execution of that idea after multiple failed attempts. He mentions several failed tablets that predated the successful iPad, explaining that these early attempts were missing critical elements. Once all the necessary components are in place, the invention moves to the marketing and business phase, where it must be presented in an appealing package to achieve success. This shift from numerous failures to one commercially viable product is typical of technological innovation.

Thought as a Complex Process:
Danny Hillis emphasizes that thought isn’t a simple process but a compilation of several algorithms and heuristics working together. He proposes that when we perceive something or experience a certain emotion, thousands of different processes are at work within the brain. These might involve associations with past experiences or various other processes, some of which might not make intuitive sense if explained.

Thought as a Summary of Processes:
According to Hillis, the actual “thought” is just a summary or headline of these intricate processes. It is our brain’s way of making sense of the numerous, and possibly contradictory, processes occurring within it.

Computational Parallel to Thought:
Similarly, when a computer, like IBM’s Watson, produces an answer, it’s not because of a single algorithm, but due to numerous processes at play. Some of these processes could be basic, such as word associations. Although these processes may seem simplistic or “stupid,” their combined action can produce something more sophisticated like an answer to a Jeopardy question.

Thinking as a Summarized Process:
Hillis argues that thinking, either in humans or computers, is a summarized output of multiple, concurrent processes. The resulting answer or “thought” is the summation of these complex processes.

Understanding Bodily Processes for Preemptive Medicine:
Hillis emphasizes the need for a deep understanding of the body’s natural processes over time to implement effective preemptive medicine. As an example, he cites HPV virus exposure leading to invasive cervical cancer in women. Despite millions of women being exposed to the virus, only a small percentage actually develop the disease over time because the body naturally sheds the virus.

Risks of Uninformed Intervention:
Hillis warns about the risks of intervening without proper understanding, turning individuals into patients unnecessarily and potentially causing harm. He stresses that the body is effective at maintaining health despite numerous possible issues.

Body’s Health Maintenance and Hidden Issues:
He points out the remarkable capacity of the body to maintain health despite numerous potential problems, some of which may be serious but are kept in check by the body. Hillis suggests that doctors usually see patients when their body can no longer maintain health, missing potential underlying issues that the body is dealing with effectively.

Striking a Balance in Preemptive Medicine:
The challenge lies in discerning when intervention is needed and when it could lead to over-treatment. Hillis suggests that the aim should be to develop a model that tracks bodily processes and identifies when the body starts to lose control of a problem. This approach would enable medical professionals to intervene effectively and prevent over-treatment.

The Complex Manifestations of Body’s Defence:
Hillis gives an example of how the body’s defense against one condition (like viral infections) can potentially lead to other health problems (like immune disorders). In such cases, while the original threat is managed, the body may manifest different symptoms due to increased immune activity. This exemplifies the complexity of the body’s health maintenance processes and the challenges in preemptive medicine.

Exploring Proteomics for Treatment Development:
An audience member expresses interest in proteomics, particularly in studying how bodies fight off infections and viruses independently. They propose the potential of observing these processes and creating models to develop drugs or treatments for individuals who may not possess this capability themselves.

Early Stages of Proteomic Exploration:
Hillis responds that the exploration of proteomics is still in its early stages, with only recent advancements allowing for the measurement of these vital variables. He acknowledges the potential of proteomics to inform medical treatments and drug development.

Optimistic Uncertainty:
Despite the uncertainty surrounding how much information can be derived from proteomic data and how it could be utilized for treatment development, Hillis remains optimistic. He perceives the exploration of proteomics as an exciting opportunity in medical science, even though it’s yet to be fully understood or utilized.

Moral Thinking in Artificial Intelligence (AI):
An audience member compares Hillis’s explanation of subconscious and conscious functional thinking to David Brooks’s division between functional and moral thinking. She asks if AI could ever enter the realm of moral thinking. Hillis asserts that moral intelligence is a critical component of intelligence and that for people to trust machines with specific types of judgments, these machines must exhibit some degree of moral intelligence.

Prescribing and Reflecting Moral Intelligence:
Hillis suggests an interesting situation where humans can prescribe the moral intelligence in AI, much like parents teach morals to their children. In trying to program machines to exhibit moral behavior, we are likely to see our moral models reflected back at us, which could lead to realizations about discrepancies between our moral instructions and the results.

Learning from AI’s Moral Intelligence:
Hillis anticipates that as we attempt to impart moral reasoning and principles to AI, we will likely discover more about our own moral systems. He speculates that our prescriptive morals might not align well with what we regard as moral behavior in practice, suggesting a potential mis-codification in our moral system. Drawing from past experiences in AI development, Hillis expects that we might learn about morality from the process of trying to instill moral behavior in machines.