By Ophelia Lee ‘26
In 2023, the “Godfather” of AI, Geoffery Hinton, stepped down from his position in Google to warn tech companies about the dangers of generative AI. Afraid of the technology he helped create, Hinton believed that AI could surpass the intelligence of humans within this generation and result in superintelligence.
Superintelligence arises from technological singularity: technology enters successive loops of self-improvement cycles in which each stage of artificial intelligence gives rise to an even more powerful entity. The end results could contain unforeseeable outcomes and, as Stephen Hawking feared, the extinction of humanity.
Typically, artificial intelligence is associated with large language models (LLMs) like ChatGPT or Google Bard, which are trained with data provided by software developers. LLMs can analyze language and generate responses based on existing knowledge. However, ethicists and developers are concerned about the implications of artificial general intelligence (AGI), sometimes called large reasoning models (LRMs). Only in their developmental stages, LRMs’ goal is to be able to adapt to wider scenarios and engage in recursive self-improvement, ultimately matching a human’s intelligence. OpenAI has released preliminary LRMs in their o1 series trained to think and perform complex reasoning. When tested, the o1 models exceeded the level of PhD students on benchmark exams in physics, biology, and chemistry.
Dr. Todd Hay, a Program Manager of Pacific Northwest National Laboratory, explains the process behind LRMs: “They do a sequential step-by-step-analysis of user questions […] A lot of the current models have been fed with data that is coming off of a human understanding of that data. Now, what they’re doing on these large reasoning models is giving them the raw data and letting those large reasoning models develop their own rulesets. They might develop rules and algorithms that we simply can’t understand […] It’s creating its own language.”
There are many ethical concerns about existing artificial intelligences. For example, since developers choose the training information, they perpetuate biases. Furthermore, AI poses privacy risks, facilitating companies to collect personal information through user inputs. This information could be bought and sold at unprecedented levels. AI can also generate deepfakes, which are images and videos depicting real or non-existent people that are difficult to discern from reality.
Dr. Sergia Hay, an instructor in the Philosophy and Religious Studies department, warns against AI’s moral judgment. She states “the AI we presently have is not good at making judgments. It will plagiarize information and makeup information in some cases. It’s drawing on stuff that’s already available to it […] I’m not going to rule it out as a potential because it’s possible that we could develop large reasoning models that could do things like ethical thinking if we give it certain moral principal inputs. It would reason from a utilitarian perspective or from a deontological perspective or something like that.”
Additionally, AI has massive environmental effects. While some are hopeful about its ability to predict climate patterns that can be beneficial for avoiding natural disasters, AI is also a contributing factor to climate change. According to the International Energy Agency, a demand using ChatGPT uses ten times more energy than a Google Search. AI also produces hazardous electronic waste, often continuing mercury and lead. To mitigate this damage, Dr. Hay suggests incentivizing eco-friendliness: “Incentives usually work better than penalizations. Penalizations sometimes can be seen as externalities where they’re just part of the operating cost of the business.”
Ultimately, creating restrictions on the boundaries of AI is an ethical dilemma. Dr. Hay stresses the importance of anticipating the consequences of technology before it becomes too late: “I think it is always important for us to think about what it is we’re intending to do […] rather than the pursuit of the tool.” For example, when cars were invented, people focused on its ability to make transportation more efficient. As we know now, cars are major emitters of pollution that contribute to climate change, which was an unintended consequence of innovation.
Dr. Hay argues for implementing an ethical task force in tech companies. “It is great that large tech firms have ethicists on staff to consult about these technologies. These ethicists can answer questions like: Should we be pursuing things? Are there some consequences that we can get hints on before they actually happen and might be a serious problem for us?”
Though the future of AI seems daunting, it is also exhilarating. Dr. Hay discusses the potential of implanting AI into physical bodies. She thinks it’s interesting that “human beings have certain types of bodies. AI, even if it’s able to use language or reason in ways that we can, [currently cannot] have the subjective experience that human beings do with their bodies. There’s something important and essential about our subjective experience in our bodies that helps us make ethical decisions in a way that maybe robots or AI may not.”
Many have sketched out lofty goals for the progression of AI, but to ensure that the fears of a tyrannical intelligence do not become a reality, ethical laws must be put in place. As the great American writer Henry Thoreau succinctly warned, “Technology is an improved means to an unimproved end.”
Read more articles like this in our Fall 2024 Issue!
