The Crossroads
By Matt Konwiser
Cross Brand Technical Leader and Field CTO, IBM

I want you to think about this carefully: How far will you go to advance your business, and what are you willing to concede?
Just before I wrote this article, I spoke with an Ethics leader and business incubator sponsor, a legal advisor, two professors, and a child-safe AI startup founder. Each one of them essentially said the same thing about aggressive AI development. “Damned if you do, damned if you don’t.”
What does that mean?
If you need to build an ultra-capable AI, you need to give it a LOT of information. Much of that information will be content which could be sensitive, confidential, licensed, and yes, inaccurate. Without all of that content, the AI couldn’t be as good of an AI. With all of that content, the AI becomes dangerous. It consumes significant resources, expands the potential to cause harm to the business, and can make more mistakes.
Electing to limit the volume of information or the complexity of the AI reduces the capabilities of the model.
Examine this with a geopolitical lens. If we restrict our most advanced models, but another nation does not, we may be “protecting” our national security or citizenry while also allowing the other nation(s) to leap ahead in their technological development. We cannot think that “we’re the only ones” who have such capabilities, which means with advancement comes competition. Competition motivates people to keep pushing, and that push can (not always) drive recklessness.
Businesses are caught in a similar bind. If they use very complex AI, it’s more likely they don’t understand it, and that it requires information you wouldn’t otherwise freely hand out. But the more you give a complex model, the greater the risk that model could be “tricked” into disclosing information. Smaller models with tighter governance will be more compliant with the rules, but without AI literacy and a well-designed system, those smaller models are also more likely to underperform or not work.
It comes to this: how far are we willing to go? Will we give an AI everything, and operate with no restrictions just to stay ahead of the competition, or will we sacrifice some level of innovation and growth to maintain a more stable, controllable pattern of use and new development?
This is the central discussion about the ethical use of AI, a conversation largely falling on deaf ears. People either don’t want to deal with it, or they simply don’t know what to do about it. In order for AI development to move forward in a way that creates balance, everyone must agree and comply. If even one person, one business, or one country doesn’t, there is no point in addressing the problem at the model development stage. If one developer continues to make a version of AI without strong ethics, that one “rogue” AI defeats the purpose of enforcing ethics for everyone else. At some point, that one will get out or others will build something to compete, and the race will put us right back where we are today.
And everyone won’t agree to shut them all down, either.
The current state of AI development has left businesses, AI makers, law makers, and society as a whole in an uncomfortable position. We’re all moving forward, awkwardly, with full knowledge of our very precarious situation. We have to be careful about just how far and how fast we go.
Even if the model makers and AI builders will not or cannot address the issue, businesses still have a voice, and a choice. No matter what is happening within a geopolitical boundary or what decisions are being made by the AI makers, businesses and their boards DO have control of what they bring into their environments and how they use it.
The CFO in particular has unique controls in this case, because they can look at the total picture. Cost of the AI systems, cost of business, carbon impact/sustainability initiatives, total return on investment, and market indicators which could provide guidance about risk of one decision over another.
There are some practical strategies that can be adopted to help get the most out of AI while maintaining your ethical compass, business priorities, and required governance.
1. Plan for innovation, build for practicality. There is no need to go big if you can get the job done with less.
2. Ensure the board’s guidance and your company’s charter are upheld. It’s important that every aspect of the business operates within the ethics and governance frameworks you developed before AI.
3. Balance build vs. buy. You do not always need the large consumer AI offerings and you can build your own independent AI systems and programs with the right talent.
4. Remember your customers. Businesses sell to humans, and humans understand humans best. As powerful as AI can be, always strike a balance between human and AI workloads.
5. Watch your costs. AI deployments can have significant hidden costs or deferred expenses. Learn from others and take a sharp look at what is being achieved with AI projects, and be frugal.
None of these suggestions are particularly novel, but in this age of AI, they are often overlooked, or dismissed as suggestions instead of tenets. It is critical that these be taken seriously, and as AI infrastructure, projects, and use cases are built out, they are considered, internalized, and used as part of the decision making process.
Of course you can’t do a few of these at all, or the rest thoroughly, if you’re not AI literate. So the first step is to use it (personally), then give it some simple work tasks. If you don’t understand why it’s doing what it’s doing, ask it to explain itself. Use those interactions to keep learning about it and exploring what it can do.
You’ll find that AI is good, and AI can do good. You just have to be responsible about deploying it and find the best ways to make it work for you.
Just before I wrote this article, I spoke with an Ethics leader and business incubator sponsor, a legal advisor, two professors, and a child-safe AI startup founder. Each one of them essentially said the same thing about aggressive AI development. “Damned if you do, damned if you don’t.”
What does that mean?
If you need to build an ultra-capable AI, you need to give it a LOT of information. Much of that information will be content which could be sensitive, confidential, licensed, and yes, inaccurate. Without all of that content, the AI couldn’t be as good of an AI. With all of that content, the AI becomes dangerous. It consumes significant resources, expands the potential to cause harm to the business, and can make more mistakes.
Electing to limit the volume of information or the complexity of the AI reduces the capabilities of the model.
Examine this with a geopolitical lens. If we restrict our most advanced models, but another nation does not, we may be “protecting” our national security or citizenry while also allowing the other nation(s) to leap ahead in their technological development. We cannot think that “we’re the only ones” who have such capabilities, which means with advancement comes competition. Competition motivates people to keep pushing, and that push can (not always) drive recklessness.
Businesses are caught in a similar bind. If they use very complex AI, it’s more likely they don’t understand it, and that it requires information you wouldn’t otherwise freely hand out. But the more you give a complex model, the greater the risk that model could be “tricked” into disclosing information. Smaller models with tighter governance will be more compliant with the rules, but without AI literacy and a well-designed system, those smaller models are also more likely to underperform or not work.
It comes to this: how far are we willing to go? Will we give an AI everything, and operate with no restrictions just to stay ahead of the competition, or will we sacrifice some level of innovation and growth to maintain a more stable, controllable pattern of use and new development?
This is the central discussion about the ethical use of AI, a conversation largely falling on deaf ears. People either don’t want to deal with it, or they simply don’t know what to do about it. In order for AI development to move forward in a way that creates balance, everyone must agree and comply. If even one person, one business, or one country doesn’t, there is no point in addressing the problem at the model development stage. If one developer continues to make a version of AI without strong ethics, that one “rogue” AI defeats the purpose of enforcing ethics for everyone else. At some point, that one will get out or others will build something to compete, and the race will put us right back where we are today.
And everyone won’t agree to shut them all down, either.
The current state of AI development has left businesses, AI makers, law makers, and society as a whole in an uncomfortable position. We’re all moving forward, awkwardly, with full knowledge of our very precarious situation. We have to be careful about just how far and how fast we go.
Even if the model makers and AI builders will not or cannot address the issue, businesses still have a voice, and a choice. No matter what is happening within a geopolitical boundary or what decisions are being made by the AI makers, businesses and their boards DO have control of what they bring into their environments and how they use it.
The CFO in particular has unique controls in this case, because they can look at the total picture. Cost of the AI systems, cost of business, carbon impact/sustainability initiatives, total return on investment, and market indicators which could provide guidance about risk of one decision over another.
There are some practical strategies that can be adopted to help get the most out of AI while maintaining your ethical compass, business priorities, and required governance.
1. Plan for innovation, build for practicality. There is no need to go big if you can get the job done with less.
2. Ensure the board’s guidance and your company’s charter are upheld. It’s important that every aspect of the business operates within the ethics and governance frameworks you developed before AI.
3. Balance build vs. buy. You do not always need the large consumer AI offerings and you can build your own independent AI systems and programs with the right talent.
4. Remember your customers. Businesses sell to humans, and humans understand humans best. As powerful as AI can be, always strike a balance between human and AI workloads.
5. Watch your costs. AI deployments can have significant hidden costs or deferred expenses. Learn from others and take a sharp look at what is being achieved with AI projects, and be frugal.
None of these suggestions are particularly novel, but in this age of AI, they are often overlooked, or dismissed as suggestions instead of tenets. It is critical that these be taken seriously, and as AI infrastructure, projects, and use cases are built out, they are considered, internalized, and used as part of the decision making process.
Of course you can’t do a few of these at all, or the rest thoroughly, if you’re not AI literate. So the first step is to use it (personally), then give it some simple work tasks. If you don’t understand why it’s doing what it’s doing, ask it to explain itself. Use those interactions to keep learning about it and exploring what it can do.
You’ll find that AI is good, and AI can do good. You just have to be responsible about deploying it and find the best ways to make it work for you.

0 Comments