Abstract: In my last article, I wrote about the double standards surrounding some AI such as Anthropic that can be deemed dangerous. Examples include Anthropic’s Mythos base model and some versions of X’s Grok AI. Why should licenses be granted to develop such Generative AI models? Are licenses even granted for such AI? Such models compromise dangerous AI. On the other hand, there are AI applications, such as in medicine and in healing Earth, that are truly helpful, useful, and needful. Examples of such AI include medical cures, timely earthquake detection to save lives, water sustainability, safe steelmaking in furnaces, better mining to save lives, and so on. So, it’s not that all AI is bad, but we must not think that all AI is good. Is it that gen AI, when combined with systems, produces the uncertainties many AI leaders have mentioned? Examples and consequences of bad and dangerous AI have been presented in [1]; hence, we must first understand what constitutes good and helpful AI and how a naïve AI can evolve into dangerous AI.
Introduction
There are many AIs that are helpful to humans, and some of those, in their development, can pose a threat to mankind. These can be seen as two sides of the same coin: AI. Helpful AI, or our well-wishers in the computing world, is one side. So far, some AI have crossed into dangerous territory, and most of them fall within the segment of AI that constitutes “generative AI.” Narrow AI has a mission, and these applications aim to fulfill that mission and achieve the goals for which the AI algorithm was designed. Generative AI (or simply, gen AI) is not explainable AI. Gen AI has posed many threats, such as indecent visualizations on X’s Grok segment and, most recently, the Mythos base model, which the company itself says can pose a threat. These were Anthropic’s own words, so Mythos was kept in the testing phase for some time until the same model, with alterations to use, was released under the product category: Fable 5.
In the documentary “AI Doc: the Apocaloptimist,” many AI leaders say they are not optimistic about how AI is developing. Some AI leaders say they are sure something will go wrong with AI developments. These include the CEOs of Anthropic and Google DeepMind, as well as the head of OpenAI, among other leaders. So, if you sow a negative thought and water it, what would you get? A sapling of negativity? Right, this is what has happened with Mythos. Why should we not control where AI is going? First, there must be a clear goal in the eyes of AI leaders; only then can they build helpful products. The AI leaders are starting with a negative thought that something will go wrong. Wait, we the people are not in a hurry.
The AI leaders are in a hurry to make more, to grab more, and yes, to innovate more. But this has become a race to capture the market faster. Stop. Think of your end goals. You are running a train at such high speed, as you said in “AI Doc: the Apocaloptimist,” not us. Stop your train and refocus on where you want to go. The same energy, if you put it into solving an Ebola-like virus, can heal the world. But you choose to develop weird generative AI models. We can live without them. This is how dangerous AI has become. Wait, refocus, find an aim, and then target the mission with the GPUs. Your aim should be just two of the following-(1) all kinds of medical research for the betterment of humanity with AI. (2) all kinds of improvements we can make to live well on planet Earth, including water, waste management, agriculture, irrigation, pesticides, and ore mining. All these are vast areas to be explored. They use designated narrow AI. Most of these AIs are helpful and predictable because they are focused. They are not like generative AI, whose outputs are often unpredictable. Adding GPUs and getting more, but without aim, is not the right strategy. In my article [1], to be specific, I was discussing gen AI, as the AI. In the next section, read more about examples of each kind of AI, the helpful AI or dangerous AI, as given in my article [1].
Illustration with Examples
Henry Ford inspired his engineers through relentless persistence and a strong belief in his vision. In Napoleon Hill’s Think and Grow Rich, Ford motivated his team to develop the new engine despite their repeated assertions that it was “impossible,” directing them to persist until they succeeded. And they did. This is how it’s made. You believe in what’s right and what you can do. If you start with a dull question mark and negative thoughts, what would you end up with? Most of these gen AI models are hit-and-trial products; be it the base work, yes, hit-and-trial is more difficult than theory. Sometimes it’s about adding more compute power; sometimes it’s about adding more data. All data on Earth would soon be consumed by AI. What would remain would be compute power. Stop, rethink, leaders! You are sowing negativity in systems, and what you are getting is negative AI applications. So, in the future, licenses must be granted to operate processors large enough to affect mankind on a large scale. In the future, the AI lead must be accountable. Saying “wrong can happen” or that the AI training is running fast is no excuse. You are competing to acquire more users, and it has become an ugly race.
Illustrations of dangerous AI have been presented in my article [1]. Mostly, these are generative AI models trained on all kinds of content available on the internet. In the absence of data categorization on the internet, gen AI can take wrong turns when traversing complex neural network paths. One way is to manually supervise the large volume of data, which seems impractical unless it’s automated using techniques discussed at the end. There are many studies on interpreting the AI brain; however, no clear outputs have been documented yet.
Even biological AI has both dangerous and helpful components. Hence, before licenses are granted to AI in the chemical and biological sciences, clear guidelines must be established. It should not be like current generative AI: develop whatever the AI makes, add more GPUs, and then see. No, this is not the way to progress further.
AI can be very helpful in saving someone’s life in a highly crowded hospital or during alerts from climate calamities, or Earth-related phenomena such as earthquakes. These aims can all be applied to the medical use of AI, whether in drug discovery or other applications, and to the care of Earth and humanity. The applications where AI is helpful are listed as follows; most of them are not gen AI:
1. Drug discovery
2. Medical procedures with robotics arms
3. Vaccine innovations
4. Disease management such as dementia
5. Water Management
6. Wastewater treatments
7. Climate calamity detection and weather alerts.
8. Steelmaking furnace
9. Document management systems
10. Image processing applications
11. Biometrics
12. Railways and travel ticket booking
13. Agriculture though life cycle
14. Food safety checks
15. Processed foods
16. Potential use in mining
17. New sources of energy
18. Innovation
19. Research speedup
20. Routing paths
21. Optimal routes for transport and travel
22. Robotics for space explorations
23. Energy management
24. Boosting power generation
25. Smart watches
26. Volcanic activity and detection
27. Automatic home appliances such as washing machine, fridge etc.
28. Garbage management
29. Helpful AI gadgets, to mention a few.
These are some of the broad categories of AI applications that are helpful. However, if engineers combine these and other applications with gen AI, there is a risk that this AI could become dangerous. Hence, we must ensure that the AI systems we build are error-free and grounded in realistic assertions. Protocols must be developed to integrate the latest AI into existing AI systems in a helpful way. If an AI leader says they are not sure what wrong can go with their AI, then stop. We don’t need their products then. They can be dangerous AI. Not just that we must stop using them, but they must be banned. We need safe, predictable AI that helps humans. AI was designed to mimic the workings of the human brain. We want AI that helps humans, not AI that is dangerous to humans. Improving compute times can speed up responses and optimize prompt time and output, but we don’t want cyber threats with your extra-high-speed AI, no. On the one hand, some leaders say they don’t know how AI can be harmful; on the other hand, they develop new models with more compute. These are the double standards I mentioned in my last article.
Conclusion and Future Work
Gen AI can be made semi-supervised to help deliver better solutions for humanity and mankind’s very purpose, while also supporting the restoration of Earth, which includes humans, flora, and fauna, as a pivotal role. However, if no improvements are made, there is no point in simply increasing processing speeds and hacking the world. To stop this, licenses must be granted for rightful use, with a clear focus on aims. If the leader of an AI company says there may be a chance of errors or of AI getting derailed, it means it could happen, and it means the leader is giving flexibility for errors to occur. This does not happen in real-world applications of other products. You make a product with a clear-cut objective: the desired outputs and the required inputs. There are no “ifs and buts.” Yes, AI may be different, but that is not how you give jobs to humans with some training. If you say AI mimics the human brain, then see how you hire an engineer in a particular stream. You have full trust in their capabilities before you hire them. So, in the future, licenses must be granted to operate processors large enough to affect mankind on a large scale. In the future, the AI lead must be accountable. Saying “wrong can happen” or that the AI training is running fast is no excuse. You are competing to acquire more users, and it has become an ugly race. If you can, with confidence, say it can’t harm mankind, well and good. Else, do not develop those AI models you doubt. Please.
References
[1] (1) (PDF) Double Standards of AI: Stop your AI Model till…
Artificial Intelligence, Computer Science & Mathematics 