How Large Language Models Can Assist People with ADHD
As anyone who has ever spoken to me for 15 seconds can surmise, I have ADHD. I've had it my whole life, but I was not diagnosed until I was 27. I didn't really start paying attention to it until I was 45, though.
ADHD is a developmental impairment of the brain’s executive functions. People with ADHD have trouble with impulse-control, focusing, and organization. About 11% of children and 5% of adults in the US have ADHD.
There’s a lot of misunderstanding about ADHD, so to be clear:
- ADHD is not a behavior disorder
- ADHD is not a mental illness
- ADHD is not a specific learning disability
ADHD is a developmental impairment of the brain’s self-management system. Learn more.
Lately I have been struggling to recover from extended burnout, which is especially common for ADHD folks. We are really, really prone to burnout because of how we have to interact with the world, and how the world interacts with us. This article from Inflow dives deeper into the whys and what to-dos. Suffice to say that what I am experiencing has a significant impact on my ability to process and organize information efficiently — and as a software developer, that’s key.
Like anyone who has a job, lately I've found myself thinking a lot about how machine learning and particularly LLMs can and are impacting our lives. It seems clear to me that these tools hold a great deal of potential to help people with ADHD and other neurodivergent folks, because of their ability to adapt information from its original presentation into new formats and modalities.
A Calculator for Words: Insights from Simon Willison
Simon Willison, a renowned software engineer and writer, elegantly captures the essence of LLMs in his blog post titled "A calculator for words." He writes:
"I like to think of language models like ChatGPT as a calculator for words. [...] A calculator for words is an incredibly powerful thing. There are so many applications of language models that fit into this calculator for words category: Summarization, Question answering, Fact extraction, Rewrites, Suggesting titles, World’s most effective thesaurus, Fun, creative, wild stuff."
This perspective underscores the multifaceted capabilities of LLMs, illuminating their potential to function not merely as tools for computation, but as versatile instruments for manipulating and understanding language.
Specifically, LLMs can be used to restructure textual information in a way that makes it more accessible to those with ADHD. For example, an LLM can take a dense technical document and break it down into shorter, more digestible segments. By dividing the text into manageable parts, ADHD individuals can more easily follow the flow of information.
Additionally, LLMs can process non-text content through helper systems, aligning it in a manner that helps individuals with ADHD understand more quickly. This might involve transcribing audio from a meeting or taking screenshots of an active desktop window at regular intervals. Such processed content can then be summarized or transformed into formats that resonate better with ADHD learning styles, and keep track of details that ADHD minds can’t remember.
Furthermore, LLMs can extend their utility by offering personalized examples, explanations, or metaphors. These tailored approaches can make complex principles or tasks more relatable, thereby reducing the cognitive load for those with ADHD.
It’s really a form of translation. The world is written for non-neurodivergent folks. These tools can translate it into the “language” I use.
Practical Implementation and Integration
The power of LLMs to assist individuals with ADHD isn’t merely theoretical; it lies in practical application and integration within existing workflows, especially in the realm of software development. Some ways I have thought about:
- Guiding the user through a task one step at a time, prompting them for information, to reduce the amount of “context” the user has to maintain. This would be especially helpful with working through times when executive function is lower – steps to recover can be initiated, and presented in a way that less likely to trigger anxiety responses.
- Examining a screenshot every 10 seconds of the user’s desktop to keep track of what the user is doing. This could help overcome working memory problems, and the dreaded “memory wipe”: a few times a day, I completely forget what I was doing, and I have to try to find the path again. I often use things like my clipboard history or screenshots taken with TimeSnapper, but it’s extra work that the computer could do for me in the background — leaving me with more energy to solve harder problems.
- Identifying and tracking potential tasks. I sometimes miss the significance of things when they are presented in the middle of a longer conversation. A personalized LLM could scan transcripts from my chat and email throughout the day, and present actionable information when appropriate.
- Adaptive presentation of information, not only in terms of formatting, but also through re-wording, re-structuring, and changing modalities, to facilitate comprehension. For some, an ad blocker extension in their browser is an essential assistive system, because so much of the web is enormously hostile and predatory to the user. ML could make that easier and more effective.
Basically it’s Chappelle’s Home Stenographer plus an LLM-powered personal assistant.
Uh oh, privacy again
So basically, these systems would record everything I do and help me filter out the important bits from the stuff that’s not as important. Essentially: conduct surveillance and analysis on myself.
Trusting that data to be handled by a third party is… a bit worrisome. Just the fact that I have ADHD is sensitive medical information. Me, I bought my own Mac Studio to run ML stuff locally, but that’s not an option for most people, and most people who would be helped by a tool like this won’t be able to self-host anytime soon.
I think there is an opportunity here for motivated individuals to either:
- Help create tools like this that are both accessible and trustworthy.
- Fund the development of these tools by motivated individuals with the needed expertise.
Practical implementation requires both creativity and technical expertise, blending the innovative potential of LLMs with the specific needs of ADHD individuals in the software development industry. But the result is a symbiotic relationship where technology empowers people to do things they can’t by themselves, and the needs of people drives the development and application of these technologies.