AI is such a strange space to work in, especially coming from the point of view I have.
I’m using it to try and build up local businesses and strengthen our economy for normal people, rather than supporting the tech oligarchs who have been pushing it.
A lot of people don’t know how to do this because they only hear the terrible things that come out of the mouths of AI CEOs, AI bros, and AI politicos. Stuff about replacing workers, comparing human needs for water to datacenters, offloading their parental responsibilities to AI, and evidence that some of them (especially the politicians) don’t even undestand what AI is.
They also don’t know the difference between things like OCR, LLMs, Machine Learning, ASR, or other components of what an end-to-end workflow may contain.
For clarity, we are mostly talking about LLMs and GenAI here (all LLMs are GenAI, not all GenAI are LLMs).
They think your options are American AI or Chinese AI, and that either way, you’re giving away your data, harming the planet, and essentially helping end the world, all for a tool that’s more or less a neat trick rather than something that can actually be useful.
On the other hand, you have a lot of people who are happy to use AI for absolutely everything and have convinced themselves AI can make the next Marvel movie. They often also don’t know much about things like open weight models, theories around folder structure, or that creating an actual movie with AI (whether of any entertainment value) wouldn’t necessarily happen in a ChatGPT conversation.
Having a nuanced take is a difficult line to walk.
I see the real benefits for our economies, especially from open weight models, and the predictions that we will eventually be able to use pretty advanced models on a normal smartphone. To me, this could be a revolution from the perspective of helping low-income people around the world start businesses, solve local problems, and detach themselves from reliance on large tech companies.
I also see the harms. I see how AI has been used to hurt women, children, the environment, people in places like Palestine and Sudan, how it has stolen from artists, and how they literally feed books into frontier models and destroy them nowadays after being told to stop pirating content.
I will not be able to cover every AI use case (or harm) here, so instead instead I want to consider 3 of the main reasons people believe AI is either good or bad, and hopefully, what I intend for you to take away, is that AI is neither. That what’s most important is us, our behaviour, and our choices.
AI is non-deterministic, but it’s also relentless in its pursuit of a well-defined goal#
One of the biggest complaints I hear from naysayers is that if you ask the same model the same question 5 times you can get 5 different answers. This is especially true for open-ended questions like “How can we improve our work methods?”
I also don’t love AI for opinionated questions. “What are 5 great restaurants in Paris” is just not something an AI model should handle. Yes, they’ll aggregate the ones that are mentioned most often by users online, but over time, we have seen how certain companies game this system with clever AEO hacks.
Here is a prompt I would use for a trip:
“Make me an itinerary for a day in Paris with 5 friends, 1 of whom is vegan, 1 of whom is gluten free, and where we have an individual budget of X for each meal. We’ll be using public transportation, but one friend needs to take frequent breaks, so stops at third spaces like parks or free tourist sites can be factored in, too. We want to leave the hotel at 10 am and be home by 9 pm.”
You will get something worthwhile from such a request.
This example will help offload some of the mental labour that planning a trip with friends can present. Both will have a different answer every time you ask; some pieces may overlap between tries, but it’s unlikely to be exactly the same. At the end of the day, there’s no “perfect” answer.
When it comes to business tasks, defining a goal is the best way to get useful output. The model may take 5 different routes to get there when you try it 5 times, but if the output meets the spec, the route taken doesn’t matter as much. In fact, it mimics human behaviour: look at some of the problems on Stack Overflow and you’ll see different approaches to solving the same problem. Some solutions are more elegant than others, and some solutions may create debt that will have to be fixed later, but if your goal is simply how to get from A-B, then you are going to get wildly different output from 5 different humans in many of the same ways.
When I say AI is “relentless”, this is why it’s been so effective recently at cracking unsolved mathematical formulas. A human gets tired eventually. An AI model will keep pushing on the same problem. This is helpful. This is something we can use.
AI is only as good as its human companion’s ability to use it#
There seem to be 3 camps of people when it comes to AI:
- Double-naysayers: Those who have applied it to tasks they are experts in, and are unimpressed. They then do not attempt to use it for things they’re not good at, because they realize they lack the expertise to verify.
- Using it one way poorly: Those who have applied it to tasks they’re experts in, are unimpressed, but take a different view on things they’re not good at. This seems to be especially pertinent in creative fields like graphic design, creative writing, etc. This also tends to affect people who were more in “management” rather than doing actual work.
- Using it one way well: Those who have applied it to tasks they are experts in, were impressed by how good the output was considering how little input was given, and have fine-tuned their working style to be able to output good quality work in a domain they understand well. Oftentimes, it’s not “solving” a problem so much as following instructions for a problem the expert regularly tackles.
I don’t know if it will ever get to the point where none of this will matter, but for now, it seems to be the case that if you learn to use it as a tool to augment your own abilities, you can do great things.
But this doesn’t mean every subject domain expert is suddenly going to be 100x more productive when using AI. Like many things, it may just not be the right tool for everyone. As someone who loves AI, I don’t love it when it’s shoved down everyone’s throats, because it’s just not made for some people, and it has developed a poor reputation because of that.
For some people, for example, working remotely is ideal, while being in-office is better for others. Tools like Slack and email are better ways to communicate, whereas for others, a phone conversation or face-to-face interaction may be more fruitful. Some people are Windows users, some people are Mac users, and the truly elite amongst us (not me, sorry) are Linux users.
We need to stop telling everyone to use AI and instead identify who is made better with AI, and who is made worse. Everyone has a contribution to make, either way, and no human is more valuable than another because of their individual skills or talents.
AI can be used for good and bad#
This is the worst part of this technology. As good as it is at solving computer bottlenecks for good reasons, it can do the same for bad ones, too. For some people, they can’t accept the possibility of the bad, so they want to shut it all down. The problems with AI, I argue, are human-made, and systemic, not inherent to the technology.
Many of us are aware of how IBM technology was used in the genocide of Jewish, Romani, LGBT, and disabled people during WW2. Their punch card system was reportedly used as part of the orchestration of the genocide.
IBM technology was also used in South Africa to enforce apartheid.
The technology itself was not inherently evil. It was not invented to be used towards evil. But, inflicting suffering on a mass scale requires organization, organization requires data orchestration, and at the time that these atrocities took place, these were the best tools for data orchestration.
Today, the best tools available, using AI, have already been implicated in the killing of schoolgirls in Minab, Iran, as well as the genocide of Palestinians in Gaza, according to Amnesty International.
Of course, the systems being used for these are more sophisticated than what you might be using on your consumer-grade laptop. The harnesses around them are connected to government intelligence, surveillance tech, and weaponry. Your computer has a browser, the Microsoft Suite, and an email inbox.
I don’t want to imply that you are bad for using something that can be used for harm. Our homes are full of items that can be used for violence, but we’re not all throwing out our kitchen knives.
I also don’t want to imply that we can go on using AI without a care or thought for how it impacts the world around us.
It’s important to recognize that we live in a society that still allows horrible things to happen. In some ways, the way our society works, especially in richer countries, requires horrible things to happen. What I do dream of is using AI to build a more equal and just world.
Here are some ways I think we can use AI for good:
- Creating smaller teams: I think AI will cause large teams to get smaller. But, I think it can also be used to make more small teams overall. Smaller businesses, sharing the pie, is better overall for society than a few large companies running everything.
- Promoting entrepreneurship overall: AI reduces the barriers to entry for entrepreneurs. You can use AI to help you understand the complexities of applying for a business permit, gathering requirements, thinking about your idea, and creating a plan for getting started. You can use it to set up your systems for marketing, your website, and so many of the other tech-based “requirements” for building a business these days. I think we’re going to see a lot more people trying something new in the next few years.
- Medical and environmental research: AI is incredibly good at crunching through large sets of data that are hard for humans to handle. With quantum computing, you can also predict problems more accurately than ever to help mitigate them. And with medical applications, I think a lot of major health problems like cancer and other diseases may be better treated or prevented in the future.
- Human rights: projects like the Human Rights Data Exchange (HRDx) that will allow human rights advocates to be able to follow emerging crises and deal with them.
In conclusion#
AI can be good or bad, effective or not effective, helpful or detrimental to society.
You could say that about a lot of things: religion, politics, cars, planes, trains.
We try to view ourselves as computers sometimes, I think. Things that operate on a set of binary rules; 1’s and 0’s. But we’re much messier than that, and the way we interact with the world around us reflects that.
I don’t think everyone needs to use AI. In fact, I think it would be a lot better if it wasn’t being shoved down everyone’s throats. There does seem to be some sort of ulterior motive for how this has all happened, and I’m sure one day people will look back and wonder how it wasn’t obvious.
But, I also love AI, and what I’m able to do with it. I try really hard to point it at computer bottlenecks, instead of human ones, and I want to help others do the same, for that list of good-idea bullet points above, and more.
If you’re interested in how you can use AI for entrepreneurship, growing your small team, orchestrating data so you can focus on your human creativity and passion, or any other positive way to use AI in a business or NGO, please get in touch.