GenAI and the Future
I’d like to share with you some of my anxieties for the future. I am a researcher in AI and my work lies at the intersection of learning and AI facilitated decision support tools. I’m also a PTA assisting in teaching at Exeter. I think often about the future of learning and the value of the skills we learn, and the transformative effect of Generative AI on them.
I share many people’s current anxieties that GenAI may increase socio-economic divides. GenAI represents a shift unprecedented since the industrial revolution. It has the potential to transform some forms of mental labour, just like the steam engine and electricity transformed manual labour. Yes, people still make a living through manual labour, performing tasks which are too difficult or diverse to automatise efficiently. However, the product of this work is rarely the focus: the value is created in the mental and artistic work that went into the product’s creation.
But the best way to learn how to design a process is through learning to make the piece yourself.
I paint miniatures in my free time. I use tools; an electric drill to help prepare the model, and an electric compressor to power my airbrush – both to paint quicker and to imitate techniques where I lack skill. The process is creative but these tools offer a shortcut.
Generative AI makes shortcuts. It can automate routine labour and transform high-level plans into specific outcomes. A hundred years ago, it used to be possible to own a lathe and make a living making a very niche product. Nowadays, large companies use CNC machines to churn engine parts by the tonne. You can’t compete yourself. You can’t match the precision or volume of machine-made parts. Due to economies of scale, a large company can employ engineers whose job is to optimise the engine design in a way that no single person can match.
As a computer scientist, I draw parallels with Software Engineering, though many others can be found. GenAI will do for Software Engineering what CNC did for manufacturing. There are few engineers who still design the necessary part, make prototypes, draw the blueprint and then programme the CNC. In Software Engineering, there will be few who design the architecture and make choices about how the programme will work. Neither group will make the product themselves. They will design the process.
But the best way to learn how to design a process is through learning to make the piece yourself. Learning requires prototyping, experimenting and understanding how the tools, knives and lathe work. The best way to learn how to make a programme is to write it. To know which libraries exist, understand these libraries, and understand the underlying software and hardware. Better understanding of the tools generally leads to better outcomes.
This is the way we currently teach people at Exeter. We teach them to understand the system, and to learn enough of the underlying principles so that they can glean the rest by programming.
I don’t know if that’s the right thing anymore. Jobs no longer expect you to write source code. Gen AI writes the code for you: you design the architecture. This labour used to be unavoidable. It’s how you historically learned the skill to move into a more senior position. I don’t think we are losing capability to make programmes, just as we have not lost the capability to make machine parts. But it feels like the distinction matters.
I mourn the loss of efficiency that came with fully understanding the program and its environment. You can no longer learn on the job. I mourn the people who have learned and prepared for a job that will no longer exist. I’m anxious about how the hiring process works. I recently heard a complaint that you are expected to use as much AI as you can during your job, but during your interview, you can’t. You are expected to know everything off the top of your head.
I’m no longer sure what the correct way to teach people is. Some research shows that Gen AI usage can alter what a person learns, sometimes irreversibly. Should we teach people how to work without AI so that they are not reliant on means of production owned by somebody else? Or should we teach them to be efficient even though it means being exploited?
There are ethical concerns. There is one big difference between the steam engine and GenAI. Each engine had to be built independently. Programmes, AI included, can simply be copied once they are created. When workers built steam engines, they were compensated for their labour for each engine they built. Like the steam engines, AI enhances the capability of capital to create value, replacing some mental labour. It removes the influence, ownership and share of labour of the worker on the product. Should we really prepare people to work under these conditions? How do you begin to do that?
I mourn the loss of efficiency that came with fully understanding the program and its environment.
Most GenAI models were trained on appropriated data. They were trained on public domain literature, unlicensed writing in public fora, and proprietary writing. Most writers could not have meaningfully consented to this, even if we are told their consent was implicit. Their labour was stolen. Should we teach people that they have no choice but to profit from this theft?
The future will require rigorous regulation of AI. Academia must be vocal and insightful about what regulation should look like. It needs to be equitable, and not to perpetuate or amplify existing inequalities. We need this despite labour displacement and relationship changes between the worker and capital. It’s a conversation we must have between disciplines. How do we collectively teach our students to be ready to contribute to those conversations?
I find myself facing these dilemmas. What and how should I teach? Is it possible to democratise access whilst relying on tools that by their nature restrict it and empower those who own them?
This article was not written or edited with GenAI.
Zdeněk Plešek
