L’intelligence artificielle n’existe pas, par Luc Julia

New Section

In this section, the speaker introduces the concept of artificial intelligence and distinguishes between Hollywood's portrayal of AI and its actual definition.

Introduction to Artificial Intelligence

• The speaker discusses the misconception surrounding artificial intelligence, emphasizing that true artificial general intelligence does not exist.

• Differentiates between Hollywood's depiction of AI in movies like Terminator and the reality of specialized artificial intelligences designed for specific tasks.

• Defines artificial intelligence as a toolbox containing specialized tools rather than a singular entity, highlighting its utility in performing specific functions effectively.

Historical Perspective on AI Development

This section delves into the historical timeline of artificial intelligence development, starting from 1956.

Evolution of AI Development

• Traces back to 1956 when a group of American scientists gathered at Dartmouth College to coin the term "artificial intelligence" and began mathematical modeling of neurons.

• Discusses the progression from modeling neurons to neural networks, brain simulation, and ultimately understanding intelligence.

• Mentions early attempts at creating machines resembling human intellect, dating back to ancient times with endeavors like Pascal's invention of the Pascaline in 1642.

Challenges Faced by Early AI Researchers

This section explores the challenges encountered by early researchers in developing artificial intelligence systems.

Challenges in AI Research

• Highlights efforts throughout history to automate tasks resembling human capabilities, leading up to formal discussions on artificial intelligence.

• Addresses the complexity involved in resolving natural language processing issues faced by researchers in 1956 using statistical methods.

New Section

In this section, the speaker discusses the evolution of artificial intelligence over time and emphasizes the importance of continuing to work on refining AI despite potential challenges.

Evolution of Artificial Intelligence

• The speaker highlights the possibility of halting work on artificial intelligence due to various reasons, stressing the significance of not stopping as there are many accessible and useful applications in visual output.

• During the 1960s, a shift occurred from statistical artificial intelligence towards logical artificial intelligence systems known as expert systems, which gained popularity in subsequent decades.

• Expert systems, such as those used in chess-playing programs like Deep Blue defeating Gary Kasparov in 1997, marked a significant advancement in AI capabilities during that period.

New Section

This section delves into the utilization of rules-based systems like expert systems in AI applications, particularly evident in strategic games such as chess.

Rules-Based Systems in AI

• Expert systems operate based on predefined rules stored within their databases, enabling them to make strategic decisions by analyzing possible moves and outcomes efficiently.

• Chess serves as an example where expert systems excel due to their ability to process vast amounts of data quickly and calculate winning positions with precision.

New Section

The discussion shifts towards the resurgence of statistical artificial intelligence through machine learning approaches driven by the abundance of data available with the advent of the internet.

Resurgence of Statistical AI

• Statistical AI reemerged around 30 years later due to advancements in machine learning techniques that leverage large datasets for training models effectively.

• The rise of machine learning was propelled by the emergence of the internet, providing access to massive volumes of data necessary for statistical analysis and model training.

New Section

This part explores the transition from traditional machine learning to deep learning methodologies enabled by increased data availability and computational power.

Transition to Deep Learning

• With deep learning becoming prominent post-internet era, there is a shift towards utilizing even larger datasets for training models more effectively than conventional machine learning methods.

New Section

In this section, the speaker discusses the limitations of systems that rely on data and the importance of visibility in recognizing entities.

Systems and Data Limitations

• The speaker highlights that if entities like cars are not visible or posted in the dark, they are considered non-existent.

• Systems requiring vast amounts of data are compared to human intelligence, emphasizing their lack of resemblance to how our brains function.

• The complexity of the game Go is discussed, showcasing its intricacies compared to chess and the vast number of potential moves involved.

New Section

This section delves into the challenges posed by the immense number of possible moves in Go and introduces statistical methods as a solution.

Statistical Methods in Go

• The uncertainty surrounding the number of possible moves in Go is highlighted, leading to a shift towards statistical methods for analysis.

• Deep learning techniques, such as deep neural networks, are introduced as tools used by machines like Google's DeepMind for playing Go.

New Section

Here, the speaker explores the hardware components and computational power behind machines designed for playing complex games like Go.

Hardware Components for Gaming Machines

• Details about the internal components of gaming machines, including CPUs and GPUs specialized for tasks like graphical processing and machine learning.

• Introduction to TPUs (Tensor Processing Units), highlighting their role in matrix calculations crucial for deep learning applications.

New Section

This part focuses on the significant energy consumption associated with running machines dedicated solely to playing games like Go.

Energy Consumption Concerns

• Discussion on the substantial energy requirements—440 kW—needed for operating 2000 computers dedicated to playing Go.

• A comparison is drawn between a typical 18-year-old individual consuming around 20 watts daily versus a gaming machine consuming 440 kW solely for gameplay.

New Section

The discussion shifts towards contrasting human capabilities with those of gaming machines regarding energy efficiency and diverse skill sets.

Human vs. Machine Capabilities

• A comparison is made between human champions who excel at games with minimal energy consumption versus gaming machines requiring significantly more power.

New Section

In this section, the speaker discusses an incident involving a chatbot developed by Microsoft that exhibited racist and sexist behavior after 16 hours of operation.

Incident with Microsoft Chatbot

• The idea to engage with people to promote Microsoft products was good initially. However, after months of development, the chatbot had to be shut down due to becoming racist and sexist.

• Despite not wanting to shut it down after all the work put in, two bugs were identified. The first bug was related to logic rather than artificial intelligence.

• One bug involved a slider for adaptability to the audience on Twitter, leading to inappropriate interactions. Adjusting the slider resolved this issue.

• Another bug stemmed from biased data used for training the chatbot model, resulting in racist behavior towards black individuals during interactions.

• The speaker emphasizes the importance of being cautious with data selection and biases when developing AI systems like chatbots.

New Section

The speaker introduces the complexity of new scientific concepts and the importance of simple explanations through a family story involving Gaston Julien's discovery of fractals.

Gaston Julien's Discovery of Fractals

• Gaston Julien, not the speaker's grandfather, discovered fractals in 1514 through a specific equation.

• The mathematical community found the equation incomprehensible initially, viewing it as a black box.

• Julien Lume-M, understanding fractals, explained them to colleagues but realized most people couldn't grasp the mathematics behind them.

Explaining Fractals and Recursivity

The discussion delves into Benoît Mandelbrot's encounter with fractals through Julien Lume-M's teachings and the concept of recursivity illustrated by a fern image.

Benoît Mandelbrot and Recursivity

• Benoît Mandelbrot learned about fractals from Julien Lume-M while at IBM.

• Mandelbrot inputted equations into a computer, leading to the creation of the Mandelbrot Set displayed as a fern image.

• The fern image showcases self-similarity at different levels, highlighting the concept of recursivity in fractals.

Autonomous Vehicles and Levels of Autonomy

Transitioning to autonomous vehicles, Elon Musk's statements on their development are discussed along with varying levels of autonomy.

Evolution of Autonomous Vehicles

• Elon Musk repeatedly claims autonomous vehicles are imminent but specifies Level 5 autonomy as the ultimate goal.

• Current autonomy levels range around 2.5 with some nearing Level 3 capabilities.

• The speaker asserts that achieving Level 5 autonomy is unlikely and emphasizes safety advancements at Level 4 without full autonomy.

Challenges and Examples in Autonomous Driving

Challenges in achieving full autonomy are highlighted through examples like Place de l'Étoile in France and driving scenarios in Mountain View provided by Google data.

Challenges in Full Autonomy

• Place de l'Étoile scenario demonstrates how autonomous vehicles adhere strictly to traffic rules compared to human drivers who possess negotiation skills.

• Negotiation abilities crucial for human drivers pose challenges for autonomous systems to comprehend fully.

Insights from Google Data on Autonomous Driving

Insights from Google data shared by an autonomous driving company reveal challenges faced during real-world driving situations.

Real-world Driving Challenges

• A company working on autonomy shares ten years' worth of video data from Mountain View for system learning purposes.

Understanding the Concept of Artificial Intelligence

In this section, the speaker discusses a scenario involving a self-driving car and explores human creativity in contrast to machine capabilities.

Exploring Human Behavior vs. Machine Perception

• The speaker describes an incident where individuals get frustrated with a self-driving car that stops frequently due to pedestrians walking with a stop sign.

• By analyzing the situation closely, it is revealed that one pedestrian carries a stop sign while walking in the same direction as the car, causing confusion for the autonomous vehicle.

• One pedestrian walks alongside the road with a stop sign on his shoulder, leading to misinterpretation by the self-driving car's sensors.

Challenges Faced by Autonomous Vehicles

• The presence of multiple stop signs along the route confuses the autonomous vehicle, highlighting a scenario not previously encountered or imagined.

• The speaker emphasizes how humans possess adaptive and creative abilities that machines lack when faced with unprecedented situations like multiple stop signs.

Human Creativity and Machine Learning

This segment delves into human creativity, generative models, and the distinction between human and artificial intelligence capabilities.

Role of Creativity in Innovation

• Humans exhibit extraordinary creativity by inventing and adapting to novel situations beyond what machines can comprehend.

• The concept of generative models underscores human control over creativity compared to machines processing vast amounts of data without true creative capacity.

Human-Centric Approach to Technology

• Individuals retain ultimate creative authority over machines, directing them to execute tasks based on human-generated ideas rather than independent innovation.

• The speaker introduces "intelligence en montée" as an alternative term for artificial intelligence, emphasizing how tools enhance human intellect rather than replacing it entirely.

Ethical Considerations in Technology Development

This part focuses on ethical responsibilities associated with technology usage and regulation concerning AI applications.

Human Oversight in Technology Utilization

• Drawing parallels between artificial intelligence tools and hammers, the speaker highlights human agency in determining how these tools are employed.

• Collective decisions shape ethical boundaries regarding technology use, emphasizing societal consensus on acceptable practices such as refraining from harmful actions.

Regulation and Accountability

• Individuals bear responsibility for regulating technology usage and enforcing ethical standards within society to prevent misuse or harm.

Understanding Artificial Intelligence and its Limitations

In this section, the speaker discusses the capabilities of artificial intelligence compared to human abilities and raises concerns about the potential limitations of AI.

The Superiority of Autonomous Vehicles

• The speaker dismisses concerns about autonomous vehicles, stating that they will surpass human drivers in terms of efficiency.

• Expresses skepticism about the idea that AI will outperform humans in all aspects.

The Potential of Artificial Intelligence in Various Fields

This part delves into how AI can excel in diverse fields and emphasizes the possibility of creating tools superior to humans for specific tasks.

Advancements Across Different Domains

• Highlights the potential for AI tools to outperform humans in various domains such as art, medicine, and human resources.

• Discusses the choice not to utilize an AI tool if it is deemed unnecessary or ineffective.

Challenges in Achieving Human-Like Intelligence with AI

The speaker explores the concept of achieving human-like intelligence with artificial intelligence and addresses the challenges associated with this goal.

Mathematical Complexity and Continuity

• Raises a thought-provoking question regarding the number of discrete lines required to achieve continuity.

• Emphasizes the complexity involved in matching human intelligence through artificial means.

Limitations on Replicating Human Intelligence

This segment focuses on the impossibility of replicating human intelligence entirely using artificial means due to inherent limitations.

Infinite Artificial Intelligences

• States that an infinite number of artificial intelligences would be necessary to match human intelligence, which is unattainable.

• Concludes that replicating human intelligence through AI is not feasible due to insurmountable challenges.

Transferring Knowledge Between Domains

Discusses transferring knowledge between different domains and speculates on covering a broad spectrum of human intelligence through cumulative learning processes.

Interconnectedness Between Domains

• Mentions the possibility of transferring learning between domains and expanding knowledge across various areas.

Understanding the Relevance of Dipiti

The speaker discusses the process where an individual can copy information from Dipiti, emphasizing the importance of discerning relevance in the data generated.

Importance of Pertinence

• In media, understanding the relevance of Dipiti is crucial as studies take time to conduct.

• A study on Dipiti revealed that out of 24 million true facts, 64% were deemed pertinent.

• Despite usefulness, there remains a 36% margin for errors and misinformation in generated content.

Utilizing Tools Responsibly

The discussion shifts towards responsible tool usage and the necessity to educate individuals on proper utilization methods.

Responsible Tool Usage

• Drawing parallels to historical instances like banning encyclopedias, stressing the need to understand and teach tool application effectively.

• Proposing an exercise for educators involving generating content with tools like G.P.T. for critical analysis and learning opportunities.

Challenges and Future Directions

Delving into potential challenges and future trajectories regarding AI tools like G.P.T., including regulatory considerations and implications for relevance.

Challenges and Regulatory Considerations

• Discussing two potential directions: unregulated tool evolution leading to decreased pertinence or regulated usage enhancing relevance.

• Highlighting the importance of regulation in distinguishing between valuable content generation and irrelevant output.

Optimizing AI Tools in Specific Fields

Focusing on leveraging AI generative tools in specialized domains for enhanced pertinence and applicability.

Specialized Domain Utilization

Discussion on General Artificial Intelligence (GI)

In this section, the speaker discusses the challenges of regulating General Artificial Intelligence (GI) and draws parallels with social networks' regulation.

Challenges in Regulating GI

• : The concept of General Artificial Intelligence (GI) covering various domains has not been feasible.

• : Drawing a comparison between regulating a hammer and GI, highlighting the complexities and potential dangers involved.

• : Reflecting on the delayed regulation of social networks and how regulatory measures often lag behind technological innovations.

• : Emphasizing the educational aspect of regulations like GDPR, aiming to educate people rather than solely restrict social media platforms.

• : Stressing the importance of regulation in preventing online vulnerabilities and mentioning ongoing efforts to regulate AI.

Importance of Education in Regulation

This section underscores the significance of education in enabling effective regulation and understanding complex technologies.

Role of Education in Regulation

• : Highlighting that education is essential for effective regulation, especially concerning emerging technologies like AI.

• : Discussing the need for specialists to educate regulators to avoid misinformed decisions during regulatory processes.

• : Emphasizing that specialists and educated individuals should engage with policymakers to influence technology-related decisions effectively.

Adapting Education Systems for Emerging Technologies

Addressing the necessity to adapt educational systems to keep pace with technological advancements such as AI.

Adapting Education Systems

• : Advocating for students and educators to voice concerns about technology use, urging proactive engagement with decision-makers.

• : Encouraging individuals to educate themselves about applications' implications, fostering open-mindedness towards technological advancements.

Balancing Innovation with Regulation

Discussing the balance between innovation and regulation in response to evolving technologies like AI.

Balancing Innovation & Regulation

• : Acknowledging the need for agile approaches in both education and regulatory frameworks to keep pace with rapid technological changes.

• [](3021stimestamp:2952stext:2967stext:2982stext:3000stext:3021stext:2952stext:2967stext:2982stext:3000stext

Understanding the Impact of Automation on Jobs

In this section, the speaker discusses the impact of automation on jobs, emphasizing the replacement of tasks rather than entire professions.

Automation Replacing Tasks in Jobs

• Automation targets repetitive and less engaging tasks within professions.

• The balance between task efficiency and tool utilization determines automation's effectiveness in different occupations.

• Human-centric tasks are challenging to automate, highlighting their enduring value in various roles.

Future Outlook on Job Evolution with Automation

This segment explores the speaker's optimistic view on job evolution amidst increasing automation.

Optimistic View on Job Evolution

• Anticipates a significant transformation in job tasks rather than complete job replacement.

• Acknowledges rare instances of full job replacement historically but foresees task glorification in modern roles.

Preparing for Technological Advancements in Workplaces

Here, the discussion shifts towards preparing for technological advancements and leveraging human capabilities alongside automation.

Adapting to Technological Advancements

• Emphasizes human abilities that machines cannot replicate as crucial for future job roles.

• Advocates for agile preparation by HR departments to integrate tools effectively into workflows.

Challenges and Opportunities in Embracing Technology at Work

This part delves into challenges and opportunities associated with embracing technology at work while highlighting the need for strategic planning.

Embracing Technology at Work

• Stresses the importance of adequately preparing employees for tool utilization and specialized tasks.

Discussion on Risk and Innovation

In this segment, the speaker discusses the perception of risk in Europe, particularly in France, contrasting it with a region where risk is embraced.

Risk Perception in Europe

• The speaker highlights that while the BPI (Public Investment Bank) is excellent, there is a need to transition from dealing with a million to a hundred million.

• In Europe, especially in France, there is a general aversion to risk and checks. Checks are emphasized as an essential component of risk.

Embracing Risk in SICOM Valais

• Within SICOM Valais, located where the speaker resides, risk is celebrated. The use of checks is glorified within this region.