How could artificial intelligence help us achieve the Sustainable Development Goals.
Back on the 29th of April I emailed Max Tegmark about his 2020 paper titled: “The role of artificial intelligence in achieving the Sustainable Development Goals”
Since it took me hours to gather my thoughts and write that email to him, I will just copy paste it here and maybe make some changes and add additional commentary whether right now, or over time as I see fit, I think that it’s still a very interesting read! Both his paper and my email to him.
I introduced myself and expressed interest in his paper, and mentioned how it aligns with my interests and passions, and then carried on saying:
Upon further research, I discovered your "The Universes of Max Tegmark" website and learned about your interests, which I share, as it states on your website:
You love big questions such as the nature of intelligence, how the brain processes information and why some but not all quark blobs are conscious.
You're interested in learning and using advanced tools from condensed matter physics, field theory or information theory.
You enjoy working with computers and state-of-the-art data to put theories to the test.
I often find myself awake at night, contemplating the questions you mentioned, as well as broader questions about our place in the universe and all of its inner workings. I have always been deeply inquisitive and curious about the world around me, eager to learn about a wide range of subjects, especially those that capture my interest, including your mentioned condensed matter physics, field theory, and information theory.
As for working with computers, my passion for computer science and technology also extends to how these fields integrate with other areas of study and research.
Now getting back on the original topic of the email, the paper. If I understand correctly, your paper's main points are the discussion of the relationship between AI and how it could benefit society today and in the future, as well as achieving the UN's SDG goals.
Going over all of the main ones with the most significant potential for positive impact first, like SDG 3, SDG 7, SDG 9 and SDG 13
I can definitely see how AI can be excellent for all of those SDGs, particularly education, as those fields exactly require a ton of objectivity. About SDG 3, looking at OpenAI's paper on GPT-4, they have it take a bunch of exams and doing very well on a vast majority, with people further having it to do exams like the USMLE, with the AI passing it. Not only that but it was able to successfully diagnose a 1 in 100,000 condition in seconds namely congenital adrenal hyperplasia, which would likely be quite a bit harder for regular doctors to do especially with such proficiency. However I have a few solutions that I would like to propose and give you that I have come up with personally after pondering on this paper and topic for an entire day.
Tackling SDG 7 and 13 at once, I would like to propose a solution that is a bit more off topic, namely we could, and should also tackle the issue of AI models consuming incredible amount of power to run like recently OpenAI stating that it cost them 700,000 dollars a day to run ChatGPT. One way I thought of doing that is that we could have AI / develop an AI to help us with innovating our semiconductor technologies, like recently Microsoft proposing a solution in the form of an AI accelerator processor to heavily optimize the costs of running an AI. Another way of doing that is we could have AI help us develop even more efficient solutions like ASICs specifically designed for running AI related workloads, like LLMs.
Direct solutions to SDG 7 and 13, reading your paper you and your co-authors proposed the integration of AI into our power grids to intelligently manage electrical demand to times when the sun is shining and the wind is blowing. With my proposals being that we could have AI present solutions to issues that are preventing us from for example turning a big part of the Saharan desert into a giant solar form to power Africa and the rest of the world with a lot more renewable and clean energy.
To solve issues like the heat radiate from the solar panels affecting the climate, efficiency of the panels and the output, we could also have AI help us advance our renewable energy technologies including solar, for more heat-absorbing and efficient solar panels that are for example integrated with a cooling mechanism like liquid cooled tubes running throughout the solar panel in a pattern, like heatpipes running through the fins of an air cooler on a CPU.
Water for such an operation could be obtained from ground water utilizing even an ancient technology like a Qanat system, which is fed into a clean and power efficient filtration and cleaning plant running off of spare power from the panels, before being fed to the panels. The panels could also be integrated with photo sensors that detect the direction of sunlight, which then feeds that info into a tiny and robust electric motor either under the panel in a big foundation, or within the panel itself which turns the panels directly towards the sun to maximize efficiency.
AI can also analyze various factors such as solar irradiance, land use, and topography to determine the best locations for solar farms in the Sahara, AI can also predict equipment failure and optimize maintenance schedules which reduces downtime and costs.
For infrastructure planning: AI can optimize the design and layout of solar farms, taking into account factors such as energy storage, transmission lines, and local weather conditions. This can help reduce the cost of infrastructure and improve the efficiency of energy production.
We could also have AI help us build a system for integrating with other renewable sources, it can help optimize the integration of solar power with other renewable energy sources such as wind or hydropower. This can help create a more stable and reliable energy supply, reducing the need for fossil fuels.
Addressing environmental concerns again with a different proposal: AI can help monitor and minimize the environmental impact of solar farms, ensuring that local ecosystems and wildlife are protected, and notify accordingly if anything was wrong.
To address SDG 13, AI can contribute to climate change mitigation and adaptation efforts. For example, AI can help improve climate models and forecasts, which can inform policies and planning for climate change adaptation. AI can also be used to optimize the allocation of resources for climate change mitigation projects, identify the most effective strategies for reducing greenhouse gas emissions, and monitor the progress of countries in achieving their climate commitments. I will address SDG 4 along with any SDGs that are at risk from emerging AI technology below.
To tackle the SDGs that are particularly at risk from emerging AI technology, namely SDG 5, SDG 8, SDG 10 and SDG 16.
Starting with SDG 5. SDG 5 is a valid concern even today with GPT unfortunately, though it should be possible to make a filtration system or perhaps even an AI that can filter out, or at least flag slurs, or any other profanity for inspection either by said AI itself, or passed along for human inspection if it can't figure it out. OpenAI already does that with the chats that people have with ChatGPT, they store it and monitor it for abuse. Plus OpenAI already utilizes both contracted and volunteer "testers" to constantly test the bot for bugs, loopholes, or any other bias, false information, or whatever else the chat bot might be trying to spread. I feel like such solutions both artificial and physical ones, can be implemented for most every AI system that has direct interactions with people.
We can already use natural language processing techniques to analyze and understand the context of text data. There are already existing solutions like content moderation APIs or AI-powered moderation tools that can help filter out offensive content. These tools can be trained to recognize and flag a wide range of inappropriate language, including slurs and profanity. The flagged content can then be reviewed by human moderators or automatically filtered out based on predefined rules. However, these systems are not perfect, and there can be false positives and negatives. Therefore, combining AI-based filtration with human oversight is often the best approach to ensure accuracy and maintain the balance between content moderation and freedom of expression. As AI technology continues to improve, it's likely that the effectiveness of these systems will improve as well.
Looking at SDG 8 and one solution I can think of proposing right now, is that yes, if anything it is inevitable that AI is going to filter out low-skilled jobs and workers, however, there is a saving grace. In the form of its MAJOR positive impacts on SDG 4. Low-skilled workers can be taught to either use, or can be freely enrolled in courses for whatever they wish to study, and have an AI teaching them about said subjects, this also extends to regular school teachers, ironically enough. People already do that today with ChatGPT even, it is incredibly easy for ChatGPT to teach basically any topic anyone could want to learn about to anyone. And to make sure that low-skilled workers have good coverage and assurance to knowledge is to perhaps make LLM's like GPT-4 or any other future ones open-source, so absolutely anyone can integrate them or use them however they want at will. Or better yet, with even more advanced AI in the future we could make a whole AI designed to be educative, this would also by proxy solve the issue with SDG 10, because since it's free and open-source, again, anyone can use it.
You might say that open-sourcing AI models can bring different issues along with it, such as potential misuse, privacy, and intellectual property issues. Regarding intellectual property related issues, there is one easy and effective solution I can think of, being royalties. Negotiate a set price with the AI companies for any and every time someone prompts any image containing their style or watermark or whatever. Another solution is to just straight up block the usage of the names of companies, specific artists, etc. but in my opinion that is pretty bad for everyone and halts the advancement of AI.
Misuse is going to come regardless, and has come with every new piece of technology, and we have always found out ways to solve them. A bit of a subjective statement rather than a proposal of objective solutions? Yes, however that doesn't change the truthfulness of the statement. It is a complicated problem to solve, and there probably is no one blanket solution for it, but rather a whole system of many interconnected ones. Even, ChatGPT is currently closed source, however people have still found ways of abusing it, and are continuing of winding ways for it, that is why OpenAI hires "testers" that can hunt for bugs or loopholes that can be used for malicious purposes as mentioned previously, and it has worked pretty well for them so far. So that could already be one solution.
Privacy has always been, and continues to be an incredibly major issue with technology as a whole currently. However for the past couple or a few years, we have started to wake up to that issue and have started working on it. OpenAI provides API access for ChatGPT which people can pay a really low sum for, to integrate ChatGPT into any application or program that they'd like, who's to say that companies that are gold standard when it comes to privacy like Signal cannot take advantage of that already right now? There's one solution for privacy issues already.
However, currently, there is a disconnect between policymakers and privacy oriented products / companies as a whole unfortunately. As unfortunately, many international and local governmental organizations think that they are above said ethics and rights of the people when it comes to privacy especially, and as a result Signal has a whole separate page on their website dedicated to publishing every government order and subpoena that they receive asking for all of their customers' private data, chat logs, etc. But since Signal doesn't keep any of them, they cannot give anything, and as a result they display the order on their website to show exactly how, as I mentioned, international / local governmental organizations think that they are above said ethics when it comes to people's right to privacy. So that is a separate issue that needs to be solved between privacy companies and governmental organizations themselves, AI has little to do with that and probably can't do much to help. At least I have yet to think of any solid proposals for it, I will attempt to think about it later when I have a lot of spare time and feel rejuvenated. But to address this disconnect, it is important to foster an open dialogue between privacy-focused companies, governmental organizations, and other stakeholders to reach a mutual understanding and establish regulations that balance the need for privacy with the requirements of security and law enforcement.
For SDG 16, I would like to propose solutions to the entire goal, and extend on your paper, starting with privacy and surveillance. I addressed privacy exclusively in my last couple of paragraphs. As for AI used in surveillance, that has been done for a long time now. You can buy a pretty cheap home security system with cameras that already uses AI powered pattern recognition to pick out and enhance video / photos of criminals. Phones, even budget ones, have utilized AI in their cameras, microphones, etc. for a long time as well. Using AI for surveillance can be tackled with problems utilizing solutions outside of AI, like recently I saw a project on hackaday that was basically a shirt you could wear that would overwhelm any camera sensor with a bunch of small infrared LED's. There is another ingenious solution that specifically targets facial recognition with a clever pattern of knitting, utilizing the Hyperface pattern. The same type of solution can be extended to pretty much everything regarding surveillance. It's important to consider both AI and non-AI solutions to protect individual privacy.
The 2nd subpoint I would like to propose solutions for under the SDG 16 umbrella is AI usage in military applications. For the military, I would like to argue that AI-powered offensive systems encourage war as much as they dissuade them, just like nuclear bombs or other WMD's. But yes, more regulation needs to be made taking into account future weapon systems that are more advanced and utilize more advanced AI systems, just like how the Nuclear Peace Treaty was made. AI in military applications has also been a thing for a long time already, take heat-seeking missile systems like the handheld Javelin, or any surface to air or surface to ship, etc. rocket system. Plus many many more. It's essential to strike a balance between the deterrent effect of AI-powered offensive systems and the potential for escalation. As mentioned previously, the Nuclear Peace Treaty can serve as a model for future agreements and regulations on AI-driven military technologies.
The 3rd and final one of my subpoints, also addressing what was mentioned in the paper, manipulation, misinformation and any potential bias or discrimination resulting from them. The same solution that I proposed for SDG 5 can be used. Making systems tailor-made to detecting such issues, as well as making existing AI more resistant to it. ChatGPT is already quite resistant to any of those problems, and has been built from the ground up to be "helpful, supportive, assistive, etc." And I managed to more or less verify that, that it's not programmed to just say it after the fact. The excellent examples are the hypotheticals I gave to it in regards to what if it were my assistant physically in real life, and then provided it with multiple and in some cases incredibly subjective and difficult dilemmas, all of which it managed to propose very effective solutions for while maintaining its ethical and safety guidelines. Impressive might I add. We must develop AI systems that are more robust against malicious inputs and harder to manipulate.
And that concludes the email! Let me know either by emailing me or in the comments down below what you think about this topic!
