Over the past decade I've written on social media about the concept of an AI Guardian, but I realize I've never put it into a blog post [4]. With the release of the LLM AIs [1] including ChatGP4, ChatGPT plugins, and recently GPT Agents (GPTs) the AI Guardian is much closer than I'd expected. So it's time to write something.
First, of course, I decided to ask ChatGPT. I've never seen the concept of an AI Guardian online, but evidently I've been looking in the wrong places. As of its April 2023 incarnation ChatGPT4 has quite a bit to say:
You: What do you know about the concept of an "AI Guardian" for special needs adults?
ChatGPT
The concept of an "AI Guardian" for special needs adults, particularly those with low IQ or autism, revolves around using artificial intelligence to provide support and assistance in various aspects of daily life. This concept is still emerging, but it has several potential applications:
Safety Monitoring: AI can be used to monitor the individual’s environment for safety hazards or emergencies, alerting caregivers or authorities when necessary.
Routine and Schedule Management: AI systems can help in managing daily routines, providing reminders for medication, appointments, and other important activities.
Behavioral Analysis and Support: For individuals with autism, AI can analyze behavior patterns to predict and mitigate potential stressors or overstimulation.
Communication Aids: AI-driven tools can assist in communication, especially for non-verbal individuals or those with speech difficulties, by interpreting signs, gestures, or other forms of non-verbal communication.
Learning and Skill Development: AI can offer personalized learning experiences and skill development activities, tailored to the individual’s needs and learning pace.
Social Interaction Facilitation: AI can help in guiding and facilitating social interactions, which can be challenging for some individuals with special needs.
Health and Wellness Monitoring: Continuous monitoring of physical health, including vital signs, sleep patterns, and physical activity, ensuring prompt response to health issues.
This concept is still in its nascent stages and raises important considerations regarding privacy, ethical use, and the need for human oversight. It's crucial to balance the benefits of AI assistance with the dignity and autonomy of the individuals being assisted. As a physician with experience in health informatics [2], you'd appreciate that while AI can provide valuable tools, it cannot fully replace the nuanced understanding and empathy provided by human caregivers.
Well, that covered the bases. I would love to know what ChatGPT4 has been reading on this topic.
In our family iPhones and narrow focus AIs like Google Maps have been both an essential tool for independence and a screen addiction curse [3]. The ability to locate our children was key to them learning to navigate by bike and bus. Mapping tools compensated for very limited navigation ability. Calendaring helped with anxiety and organization. Notes and Reminders helped with memory and recall limitations. Beyond the obvious advantages of messaging we discovered both our children were better able to process text messages than verbal communication -- especially for difficult topics.
Now there is the near term prospect of far more support going forward. Monitoring email, messaging and apps for the scams and frauds that live on the cognitively disadvantaged. Monitoring financial transactions for worrisome trends and events. Monitoring home environments for worsening or dangerous conditions. Encouraging better diet and activity. Acting as a job coach. Appropriately paced and reinforced lifelong learning. Guiding our children through the vast complexity of medical care. Even providing a form a companionship in time or facilitating finding company in the physical world. Doing, in other words, what Emily and I do after we are incapacitated or (inevitably) dead.
In the absence of personal immortality for Emily and I the AI Guardian is the best hope for our children's longterm health and happiness. Even as I have profound concerns about the disruption that comes with the rise of the modern AI I also have hope for a better options for the cognitively disadvantaged.
- fn -
[1] Some dislike applying the term "AI" to Large Language Models because they don't appear to be conscious or sentient. Much of my professional career was loosely related to precursors of modern AI and there is absolutely no doubt that in 1995 we'd have called ChatGPT a bloody scary AI. The goalposts are mobile. LLMs are AIs by any reasonable definition. Insofar as we understand human cognition we do much of the same things that LLMs do.
[2] This "as a physician" bit startled me until I remembered that I'd given ChatGPT4 that information as a persistent context for our sessions.
[3] I tried writing a book on how best to balance these harms but I gave up when I finally accepted that there was absolutely no interest in such a book from the community I was writing for.
[4] A 2015 post on Google Inbox was the closest I could find on the topic of AI assistance. I tagged that nsAI for non-sentient AI.
