Can AI predict the 3rd order effects of its own intervention? - DT

• Robert DT on twitter: @DeeperThrill
• Doctorate on biomedical engineering with a focus on AI
• Entrepreneur building biomedical systems with AI specifically; medical imaging

The conversation centers on the role of artificial intelligence (AI) in medical imaging, with an emphasis on computer vision and the utilization of existing imaging algorithms. Transformers, a type of deep learning model, are discussed for their unique self-attention mechanism and applications in natural language processing and computer vision.

The talk pivots to data cleaning, specifically anonymization and safeguarding personal identifiers in the context of healthcare. Questions arise about data storage in healthcare facilities and the process of transferring it to the cloud.

The conversation broadens to encompass AI's predictive capabilities and inherent risks, including the possibility of AI predicting third-order effects of its own interventions and concerns about excessive trust in AI predictions.

The potential of AI in genetic engineering surfaces, particularly regarding CRISPR technology and nanobots. The conversation explores the benefits and risks of such advancements, including the revival of extinct plants and emergence of new diseases.

Finally, the conversation shifts to societal implications of AI, including job displacement, the emergence of an attention economy, and the prospects of decentralized AI. The importance of understanding the limits of AI is underscored.

 

Show notes

1. We need to examine what's currently happening in the field of AI, particularly in relation to medical imaging. This involves an exploration of computer vision technologies and how pre-existing imaging algorithms are being applied.

2. We should discuss the concept of a "transformer" in the context of artificial intelligence.

3. A critical part of working with AI is data cleaning. This includes the process of anonymization, ensuring that we only use the person's image and not any identifiable data like their name. We must also consider the storage of this data, which is typically housed on hospital servers. Additionally, there's the question of how this data is transferred to a cloud system for further processing.

4. Let's explore the issue of gatekeeping in the field of AI. This might involve discussing the role of clinical trials and the Institutional Review Board in ensuring ethical standards.

5. The engineering aspect of gatekeeping also requires attention, particularly when dealing with 3D data sets for imaging.

6. We should highlight two major changes currently happening in the field of AI.

7. Swin Transformers represent a significant development, as they are built off the concept of transformers in AI.

8. Let's delve into the world of language modeling and chatbots.

9. We must also consider the potential downsides of these AI technologies.

10. The transhumanism angle presents an interesting point of discussion, particularly in relation to the next generation of technology.

11. For example, the development of the mRNA vaccine was a major leap forward in response to global health crises.

12. There's also the concept of generative mRNA vaccines, which use AI to generate potential cures.

13. However, these AI technologies also come with risks. They could inadvertently create a disease, or develop a cure that isn't effective.

14. The ease with which technology can be used in this field means that virtually anyone can make implants, leading to a new set of challenges.

15. We should also discuss the emerging role of AI in lab-based work, such as managing petri dishes.

16. The application of Hegelian principles to AI provides an interesting philosophical perspective.

17. Looking ahead, we might consider what a lab kit might look like in ten years.

18. The idea of the first version of something, and its relationship to anti-authoritarianism, is another interesting topic to explore.

19. We have to acknowledge that AI, despite its potential, will not prevent all risks.

20. AI can be used as a predictive tool for triaging, helping to determine whether an intervention will benefit a person.

21. The use of CRISPR technology is another relevant point of discussion, especially considering its potential downsides, its application in nanobot technology, its use in regrowing extinct plants, the potential for new diseases arising from its use, and the systematics of finding new plant species in places like the Amazon.

22. Let's also consider the case of the dodo and the role of technology in its extinction.

23. With a small sample size, AI can predict certain outcomes, a feature that can be beneficial in various fields.

24. Most plant species are discovered rather than created, and AI can potentially help in predicting where these new species might be found.

25. The question arises: is AI better at predicting the future? It can certainly help us see larger scale patterns that we aren't aware of.

26. However, the act of predicting the future can create its own issues, akin to the Oracle of Delphi dilemma. For instance, can AI predict the third-order effects of its own intervention?

27. By revealing patterns, AI becomes a more effective tool. The more layers of patterns it can show us, the better.

28. AI and Medical Imaging: AI is increasingly being used in medical imaging, particularly through deep learning techniques. These have applications in MRI, CT, and PET scans, enhancing image reconstruction, quality, and efficiency. While impressive progress has been made, the technology still needs further development before it can be widely applied in clinical settings
29. Transformers: Introduced in 2017, transformers are a type of deep learning model used primarily in natural language processing and computer vision. They're distinguished by their use of self-attention, enabling them to process the entire input data all at once, rather than sequentially as in Recurrent Neural Networks (RNNs). This allows for more parallelization and thus reduces training times. Transformers have become the model of choice for many NLP problems, replacing RNN models such as long short-term memory (LSTM)