AI – Next Steps

2023

Chon, Kilnam

전길남

KAIST

​

2023.5.17rev12.25

Contents

1. AI History

2. Generative AI

2.1 Generative AI

2.2 Open AI

3. Future of Life Institute & Center for AI Safety

4. Artificial General Intelligence

5. Europe

6. Asia

7. Selective Articles and Videos

8. Issues

9. Remarks

References

​

​

​

2

1. AI History

First Boom – 1960s

Second Boom – 1980s

Third Boom – 2010s~Present(2020s)

Part 1 (2010s): Deep Learning/Machine Learning

Deep Mind,…

Part 2 (2020s): Generative AI

Open AI,…

Remark:

3

2. Generative AI (and Discriminative AI)

2.1 Generative AI

2.2 Open AI

4

2.1 Generative AI (with Large Language Model)

• Generative AI is popular now, especially with Large Language Model.
• Multiple media; text, audio, music, image, and code.
• Taking advantage of recent advances on computing power to work on “trillion”.
• Need large investment; dominated by major companies in USA (Google, Microsoft/Open AI, Meta, and Amazon) and China (Alibaba, Tencents, Baido, and Huawei).

5

2.2 Open AI

​

• First major generative AI application products for end-users
• Open AI covers 3 major areas; Text, Image, and Audio.
• GPT-1, 2, 3, 4, 5
• ChatGPT(for text)
• DALL.E (for Image)
• Runway (for Video)
• SongR (for Music)

6

3. Future of Life Institute & Center for AI Safety

• Future of Life Institute (FLI) Workshops on AI (with AI Principle), 2010s

​

• Future of Life Institute Open Letter on the 6-month moratorium on Giant AI Experiments [Tegmark 2023]

​

• Center for AI Safety, Statement on AI Risk, “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”[CAIS]
• “AI Safety – Ethics & Society” course, aisafetybook.com/course-materials

​

7

4. Artificial General Intelligence

What if AI could be smarter than human?

​

- AGI may be arriving earlier than what we predicted.

- Shane Legg named AGI in early 2000s.

- Koch elaborate on GPT-4 and AGI in IEEE Spectrum article, 2023.

- Bubeck (MS Res.); “Spark of AGI” (consciousness, common sense,..)

- Hendriks; “Natural selection favors AIs over humans.

- Potential social disservices/disruption by A(G)I + SNS.

- Darren McKee, Uncontrollable; book, You Tube, 2023/2024.

8

5. Europe

• AI Act by EC
• Many AI leaders are European working in North America.
• UK Government organized AI Summit in 2023.
• UK is leading AI Research & Development with Deep Mind and others.

​

9

6. Asia

• Mostly China; Generative AI (Baidu, Alibaba,…), Xiaolce,
• Regulations in China [ChinAI #220]

10

7. Selective Articles and Videos

• Open AI (Sam Altman, Illya Suskever, GPT-4 Tech Report,…), 2023.
• Sevastien Bubeck, Sparks of AI, Microsoft Research, 2023.
• EU, AI Act, 2023.
• Lex Fridman - Interview; Altman, Sutskever, Tegmark, Hassabis, Bach,…
• Yuval Harrari, AI and Future of Humanity, Frontiers Forum, 2023.
• Dan Hendricks, An overview of catastrophic AI risks, 2023.
• Dennis Hassabis, Alpha*, 2010s and 2020s.
• Geoffrey Hinton, Possible end of humanity from AI?, 2023.5
• IEEE Spectrum (Koch (in GPT-4, AGI….), 2023.
• McKinsey, Economic impact of generative AI, 2023.
• Stuart Russell, Beyond Chatbot…, 2023.3, Human Compatible, 2019,….

​

​

​

​

11

8. Issues

- Regulate like nuclear technology, biotech (medicine), and/or aviation?

• Regulations (Altman, Tegmark, China, Europe/AI Act, USA,…)
• ChatGPT(and some others) is “rote”, learn by repetition, not by understaning (like 9*9)[Brooks]
• Nuclear tech. cannot produce more advanced nuclear, but AI could.
• Common sense, consciousness, intimacy,..

​

​

12

9. Remarks

• Leading AI experts: mostly European in USA.
• Need to be good at English.
• Asia is invisible except China.
• Social media with even primitive AI could disrupt society.[Harrari]
• Printing press does not produce books, but AI does.[Harrari]
• AI danger may come before climate crisis.
• Large language model (with ChatGPT) produced 100M users in 2 months.
• “AI” may be outgrowing “Digital” which includes computers, internet, and data.
• (Generative) AI: $40B/2022, $1.3T/2032 [Bloomberg/IDC/Aivalley.ai, 2023].
• Dan Hendrycks, Natural selection favors AIs over humans, 2023.
• Center for AI Safety, Statement on AI risk, 2023.(with 1000s signatories)
• (I plan to write Retrospective of Asia Internet History, 5^th Decade with generative AI.)

​

13

References

[Altman 2023] Sam Altman, Lex Fridman #367, 2023.�[Bengio 2023] Y. Bengio and J. Harrari, AI, democracy & future of civilization, 2023.6.�[Bloomberg 2023] Bloomberg, ChatGPT to fuel $1.3T AI Market by 2032, 2023.�[Bostrom 2014] Nick Bostrom, Superintelligence, 2014.�[Brenner 2023] Ian Brenner & MustafaSuleyman, Foreign Affair Interview, 2023.9�[Brooks 2023] Rodney Brooks, Just calm down about GPT-4…, IEEE Spectrum, 2023.�[Bubeck 2023] Sevastien Bubeck, et al., Spark of AI, Microsoft, YouTube, 2023.�[CAIS 2023] Center for AI Safety, Statement on AI Risk, 2023.�[ChinAI 2023] ChinAI, Generative AI, Regulation,.., ChinAI #220, 2023.�[ChinAI 2023b] Hu Hong, Regarding risks of AI, global regulation takes 1^st step, #245.�[ChinAI 2023c] Blue Paper, Large model governance, #245.�[Christiano 2023] Paul Christiano, Alignment Research Center, YouTube, 2023.�[CNN 2023] Fareed Zakaria, GPS (Schmidt, Hinton, Art), 2023.9.3�[EU 2023] EU, AI Act, 2023.[draft]�[ExMachina 2012] ExMachina, Netflix, 2015.�

14

References (continued)

[FLI 2023] Future of Life Institute (FLI), Pause giant AI experiments, Open Letter, 2023.�[Fridman 2023] Lex Fridman, Podcast Series, 2000~Present.�[Harari 2023] Harrari, AI and future of humanity, Frontier Forum, 2023.5.�[Harari 2023b] Harari, Lex Fridman #390, 2023.7.�[Hassabis 2022] Demis Hassabis, Lex Fridman #299, 2022.�[Hendrycks 2023] Dan Hendrycks, Natural selection favors Ais over humans, 2023.�[Hendrycks 2023b] Dan Hendrycks, An overview of catastrophic AI risks, 2023�[Hinton 2023] Geoffrey Hinton, Two paths to intelligence, CSER, Cambridge, 2023.6.�[Humane 2023] Center for Humane Technology, AI Dilemma, 2023.3.9.��

15

References (continued3)

[IGF 2023] IGF, Ethical principle for the use of AI in security, #33, 2023.10.11.�[ITA 2023] Int’l Trade Association, AI adds $15T to the global economy by 2030.�[Legg 2023] Shane Legg, Path to AGI, YouTube, 2023.(also AGI by 2028)�[Li 2023] Fei-Fei Li, Rise of AI Superpowers, 2023.�[Li 2023b] Fei-Fei Li, World I see, 2023.

[Marcus 2023] Gary Marcus, Rebooting AI – Building AI we can trust, 2019.�[Markets 2023] Markets and Markets, Generative AI Market, 2023.�[McKinsey 2023] McKinsey, Economic impact of generative AI, 2023.�[OpenAI 2023] OpenAI.org, Frontier Forum Forum, 2023.�[Pew 2023] Pew Res. Center, As AI spreads,….digital life by 2035.�[Russell 2023] Stuart Russell,Beyond Chatbot, Commonwealth Club, 2023.�[Social 2020] Social Dilemma, Netflix, 2020.�[Spectrum 2023] Spectrum, GPT-4 ups the ante in AI arms race, IEEE, 2023.�[Spectrum 2023b] Spectrum, GPT-4, AGI, & hunt for superintelligence,IEEE, 2023.4.19.�[Spectrum 2023c] Spectrum, The who, where and how to regulating AI, 2023.6.14.�[Stanford 2023] Stanford HAI, Generative AI, perspective from Stanford HAI, 2023.�

16

References (continued)

[Suleyman 2023] M. Suleyman & Harrari, 2023.9�[Sutskever 2023] Ilya Suskever, Eye on AI, 2023.�[TAUVOD 2023] Chat with Open AI CEO and CTO, TAUVOD, 2023.6.�[Time 2023] Time Magazine, 100 on AI, 2023.9.�[Tegmark 2023] Max Tegmark, Open Letter, 2023.�[UK 2023] UK AI Safety Summit, 2023.11.1-2.�[WhiteHouse 2023] Executive order on Safe, Secure and Trustworthy AI, 2023.10.�[Williams 2023] Danielle Williams, History of AI (diagram),daniellewilliams.com, 2023.�[Wolfram 2023] Stephen Wolfram, What is ChatGPT doing and why does it work?�[xAI 2023] xAI, 2023.�[Yudkowsky 2023] E. Yudkowsky, Dangers of AI, Lex Fridman #368, 2023.�[Zemmel 2023] R. Zemmel, McKinsey guide to outcompeting in age of digital & AI.

​

17

Abbreviations

AGI Artificial General Intelligence�A I Artificial Intelligence�GPT Generative Pre-trained Transformer�LLM Large Language Model

18