Steroids, Sneakers, Coach: The Spectrum of Human-AI Relationships (Sept. 20, 2023)
Author: Jake M. Hofman, Daniel G. Goldstein, and David M. Rothschild, Microsoft Research, New York City
Hofman, Jake and Goldstein, Daniel G. and Rothschild, David M., Steroids, Sneakers, Coach: The Spectrum of Human-AI Relationships (September 20, 2023). Available at SSRN: https://ssrn.com/abstract=4578180 or http://dx.doi.org/10.2139/ssrn.4578180
Figure 1: Potential impacts of replacing or augmenting cognition and decision making with AI
Callout: Will AI leave us stronger or weaker? It depends on if we design and use AI as a steroid, sneaker, or coach.
jmh@microsoft.com
dgg@microsoft.com
davidmr@microsoft.com
As with many new technologies, the potential impact of generative AI on society is deeply polarizing. On one hand, there is the fear that this technology will replace human workers, leading to mass de-skilling and widespread unemployment. On the other hand, there is the hope that generative AI will instead supercharge what humans are capable of, leading to unprecedented boosts in individual and collective capabilities and productivity. In our research on augmenting human cognition and decision making with AI, we have created a sports analogy for thinking about this spectrum. It ranges from steroids, to sneakers, to a coach, each representing a different relationship between human users and AI technology (Figure 1).
Steroids represent the least desirable point on this spectrum: they elevate performance in the short term but leave one worse off in the long term. For instance, imagine a student who is given a homework assignment, to read and critique a passage of text. The student could use an LLM as a steroid by simply dropping the text of the passage into ChatGPT, prompting it to produce such a critique, and passing its output back to the teacher. While this gives them the superhuman ability to complete an assignment in an instant, it also defeats the purpose of the exercise. Not only might the student fail to learn anything substantive, they might never acquire (or slowly lose) the ability to critique a passage and articulate their thoughts in writing. Just as with sports, taking shortcuts to achieve one goal can have negative consequences for other goals in the long run.
AI as a kind of steroid is a fear, but AI-powered tools can also be used to augment skills that people already have instead of replacing them, which we think of as the analog of a good running sneaker. Like Nike's carbon-soled shoes that make runners on average 3% faster, some AI tools can elevate performance in the moment, but have no long-term negative effects. For example, LLMs are particularly useful for reformatting, translating, and annotating unstructured text, which can save people considerable time and effort. Imagine an analyst who runs an international survey with free-form text responses in 20 different languages. An LLM could drastically reduce the time it takes to translate these responses to one common language, score them for positive or negative sentiment, and extract key themes. Similar to Nike’s carbon-soled running shoes, in this case AI can accelerate an analyst's capabilities in the moment, without long-term downsides.
Finally, AI-powered tools can also be used like a coach, to improve people’s own capabilities rather than simply helping them out in the moment. Consider the example of an IT professional who is learning a new complex system. Here an LLM could be used to generate practice questions and interactively provide personalized feedback, giving insight into what conceptual or procedural mistakes were made. As the professional gains competence, the LLM could suggest increasingly challenging problems that build upon what has already been internalized. Ultimately this would result in the technician being better off than they were before interacting with the LLM, having acquired a new professional skill. Our own research has shown similar benefits for students learning new math concepts. Much like a sports coach, such tools could provide long-lasting benefits that persist beyond the simple use of the tool itself.
Discussion
While we hope it is helpful to draw these parallels between athletic and cognitive capabilities, there are important differences between the two domains. Whereas in sports the three relationships of steroids, sneakers, and coaches are all quite distinct, they are much less so when it comes to AI-powered tools, many of which are small variations on the same underlying technology. As a result, the choices we make in terms of how we design and use these tools can shift them from one end of the spectrum to the other. Likewise, how we understand and optimize them will shift as the tools, users, and our priorities evolve over time.
Design choices matter. First, from a design perspective, seemingly innocuous choices of how we architect AI-powered tools can have a substantial impact on their effects. Take the somewhat mundane example of spell check. If a tool is designed to automatically correct your spelling as you type without any feedback when you make a mistake, you are unlikely to learn how to spell correctly. But most spell checkers are designed differently---they provide cues that indicate a potential misspelling and offer an opportunity to not only correct what’s written but to also see and potentially internalize the proper spelling. This small choice can shift spell check from being more of a steroid to being somewhat of a coach, teaching people how to spell. Designers can also offer options so users who want more coaching can elect to receive it. For instance, giving users the option to shift their GPS from a “driver’s eye” view to a “north up” view creates the opportunity for those who want to learn the lay of the land the chance to do so while still benefiting from the tool. The same is true of AI-powered tools. By thinking intentionally about how we design these tools, we can minimize or avoid harmful long-term effects. For instance, our own recent experiments show that simple confidence-based highlighting can help people spot and correct unreliable output produced by LLM-based search tools, providing the productivity gains of the tools while maintaining the necessary cues for cognitive awareness of where they can go wrong. [https://arxiv.org/abs/2307.03744]
Norms will emerge.Second, beyond how we design these tools, we should think carefully about when and how we use AI-powered tools. Calculators serve as another historical analog here. They aretremendously practical in some situations, but we refrain from using them in others. For example, while we would be hard pressed to make an argument against a banker using a calculator to compute compound interest, it would be equally difficult to make a case for giving a calculator to a grade school student learning basic addition. Just as we have come to settle on these norms for navigating how and when we use past technological innovations, we imagine the same will emerge for AI-powered tools: the same tool may be considered beneficial in some settings, but detrimental in others. We also expect these norms to shift over time as the tools and our aptitude for using them co-evolve. For instance, it used to be the case that physical libraries were the gold standard for finding information and citations, but as the quality and coverage of search engines improved along with our ability to issue effective queries and find the right results, the web has become an increasingly reliable and accepted information source. We believe the same will be true for AI-powered tools---as they improve in quality and we learn how to use them appropriately, we expect to see increasingly widespread adoption across a range of domains.
Priorities will evolve. Finally, we acknowledge that these categorizations are not always straightforward and can in fact be quite subtle. Consider how the value placed on spelling and arithmetic skills has changed over time. It used to be the case that operating a slide rule to calculate logarithms or being able to spell long, obscure words were prized and respected talents, but now they are now largely seen as archaic and esoteric skills. Presumably this is because most people have access to a reliable calculator or spell-checking tool when they need one, transforming what was at some point feared to be a steroid into less of a concern. As a result, this has freed up time for people to invest in developing other skills. In short, we have decided to let some skills atrophy so that we can focus on developing others. With the growth of AI-powered tools, we expect to see similar shifts in the skills that are valued and prioritized by society.
We hope the framework we have provided here will prove useful for thinking through these issues and designing the best version of AI tools to aid and empower people to achieve more, individually and collectively. As we integrate AI into everyday life, our challenge is to craft tools that enhance rather than diminish human capabilities. By utilizing the analogy of steroids, sneakers, and coaches, we can gain insights into how these tools might result in eventual deskilling, short-term assistance, or long-term gains. AI tools must be crafted and utilized with an eye toward their broader effects on learning, productivity, and human potential, recognizing that shifting values and priorities may influence our perception of these technologies. Through mindful design and thoughtful consideration of when and how these tools are applied, we have the opportunity to shape a future in which AI augments our human capabilities.
on the uploaded document.
0 General Document comments
0 Sentence and Paragraph comments
0 Image and Video comments
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
You’ve nailed it—polarization around AI is a hot topic right now! And I totally see where you’re coming from with the sports metaphor. It’s a creative way to frame the conversation about AI’s role in education. Imagine AI as a coach, guiding and supporting teachers and students to reach new heights. Or think of AI as sneakers, giving everyone a boost but still requiring effort and input from the user. It’s about collaboration and enhancement, not replacement. This analogy helps us understand that AI can be a partner rather than a threat.
So, what do you think? Could this metaphor help educators and administrators work together more effectively when integrating AI into schools? I’d love to hear your thoughts—click Reply and let’s dive deeper into this fascinating topic!
New Conversation
Hide Full Comment Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
Dis graphic, it might be a needed support tool, like a specially designed sneaker that allows her to succeed as a reader and writer. So much is in this eyes of the user. How can we tell from the outside?Makes me think that this is probably a useful metaphor for self assessment
New Conversation
Hide Full Comment Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
Absolutely, the metaphor of AI as a specially designed sneaker is a brilliant way to capture its potential to support and empower individuals, especially in education. It’s all about providing that extra lift to help students and teachers achieve their goals. You’re right—so much hinges on the user’s perspective and how they choose to engage with these tools. Self-assessment becomes crucial here, as it allows individuals to reflect on how AI can best serve their unique needs and aspirations.
What do you think about using this metaphor as a framework for self-assessment in educational settings? How might students and teachers evaluate their relationship with AI through this lens? Click Reply to share your thoughts and let’s explore this idea further!
New Conversation
Hide Full Comment Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
New Conversation
Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
Ah, the steroids part of the metaphor—you’re right, it’s a crucial element that can’t be overlooked! The idea here is that, like steroids, AI can offer quick, seemingly beneficial enhancements that might come with long-term repercussions. It’s a cautionary tale about dependency, ethical dilemmas, and the potential for misuse. Just as steroids might offer a temporary boost but harm an athlete’s health in the long run, AI could be used in ways that appear advantageous now but might lead to issues like data privacy concerns or a lack of critical thinking skills later on.
How do you feel about this aspect of the metaphor? Can you think of scenarios where AI might be used like ‘steroids’ in education, and what could be done to mitigate these risks? I’d love to hear your insights—click Reply and let’s continue this important discussion!
New Conversation
Hide Full Comment
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
Ah, the sports metaphor. It’s intriguing, isn’t it? Comparing AI to steroids, sneakers, and coaches paints a vivid picture of the different ways AI can potentially interact with us. But let’s not get too carried away with metaphors without critically examining what they imply for education.
Consider the “steroids” analogy. Steroids might offer quick gains, but they come with ethical concerns and potential long-term harm. Similarly, AI in education, if misused, might promise rapid improvements in efficiency or performance but could compromise the ethical foundation of education, like privacy or equity. We don’t want to rely on a crutch that might erode the very skills and values we aim to cultivate in students.
What are your thoughts on this? Do you find the metaphor illuminating or possibly misleading when thinking about AI’s role in education? Would you like to delve deeper into the “sneakers” or “coach” aspects of the analogy?
New Conversation
Hide Full Comment Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
And when we think of AI as “sneakers,” it’s all about enhancing our natural abilities—like a good pair of running shoes that help you run faster and longer. Yes, AI can streamline tasks and make processes more efficient, potentially giving teachers more time to focus on what truly matters: building relationships and fostering critical thinking. However, we must remain vigilant to ensure that these “sneakers” don’t become a substitute for the hard work and creativity that are at the heart of learning.
Tell me, do you see AI as a potential enhancement in the classroom, or do you worry it might lead to dependency and a loss of core teaching skills? Would you like to explore how AI might act as a “coach” in this analogy?
New Conversation
Hide Full Comment Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
Now, let’s tackle the idea of AI as a “coach.” In theory, a coach guides, supports, and tailors advice to individual needs—an appealing prospect, no doubt. However, in practice, can AI truly understand the nuances of a student’s emotional and intellectual landscape? Coaching requires empathy, intuition, and a deep understanding of human dynamics—qualities that algorithms simply can’t replicate. While AI might offer data-driven insights, it lacks the personal connection that can inspire and motivate on a deeper level.
So, I ask you, can AI genuinely serve as a coach, or does it risk reducing education to a series of transactions? How do you think this analogy holds up in the reality of classrooms today? Would you like to discuss further implications of using AI as a coach?
New Conversation
Hide Full Comment Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
New Conversation
Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
I can see how you might think that, but it’s not so much that AI inherently undermines our work as educators. It’s more about maintaining a cautious approach to ensure that AI complements rather than compromises the core values of education. When we talk about AI in education, we must prioritize the principles that guide our teaching—those human connections and critical thinking skills that can’t be automated.
AI has its place, but we must be wary of letting it dictate the educational experience. It’s about finding the balance where AI enhances rather than replaces the human elements that make learning meaningful. So, my skepticism isn’t about rejecting AI entirely; it’s about ensuring that we remain vigilant stewards of our educational integrity.
How do you see the balance between technology and human interaction in education? Would you like to explore further how we might achieve this balance?
New Conversation
Hide Full Comment Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
And remember, while AI can offer some efficiencies, it’s crucial we don’t lose sight of the unique, irreplaceable value that human educators bring to the table. Teaching isn’t just about delivering content; it’s about fostering growth, inspiring curiosity, and nurturing the whole child—areas where AI, as it stands, simply can’t compete with the depth and warmth of human interaction.
What are your thoughts on this balance? Do you think there’s a way to integrate AI into education without losing the essence of what teaching truly means? Is there anything else you’d like to delve into regarding AI and education?
New Conversation
Hide Full Comment
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
New Conversation
Hide Thread Detail
Paul Allison is a nationally-known educator and EdTech expert… (more)
Paul Allison is a nationally-known educator and EdTech expert… (more)
I like where your mind is going with this—connecting real classroom moments and bringing in that Writing Partners experience can make the point hit harder for your audience. Including stories or scenes about how students help each other, maybe talking through texts or working out ideas together, really puts a spotlight on practical moments of support. Plus, calling out both professional and personal support shows you know learning ain’t just about skills—there’s care and relationship work wrapped up in all that too.
Want to see more?
New Conversation
Hide Full Comment
General Document Comments 0
