Despite the ease that Artificial Intelligence makes for us, ethical AI for business is crucial. The case for robots going rogue may seem like a scene from a science fiction movie. But, cases where AI fool humans are springing up nonetheless.
Consider Eugene Goostman, for example.
In 2014, it (yes, Eugene Goostman is a chatbot) won the Turing Challenge – turning out to be the first robot to fool almost half of the human raters into thinking that it was human.
Essentially, a Turing test involves human raters chatting with unknown entities to study their text input to determine whether it comes from a human or AI.
To some, Eugene Goostman’s victory might read as progress in the quest to create human-like robots. To others, it signals how far along robots have come in fooling humans – a threat that raises the case for ethical AI for business.
Keeping this in mind, we wanted to discusses exactly what ethical Artificial Intelligence is, what are the ethical dilemmas associated with AI, and how to build an ethical AI framework for your organization.
Let’s dive in.
How to build ethical AI practices in your business
Being transparent about data collection with your audience, taking conscious steps to understand how AI works, and ensuring confirmation bias is limited. These are just some recommendations we’ll make today.
To fully understand these ethical AI business practices, you need to get to grips with why such ethics are essential and the moral dilemmas that Artificial Intelligence poses.
But, first, let’s quickly refresh what ethical AI is.
What do we mean by AI ethics vs. ethical AI?
To begin with, let’s clarify the difference between AI ethics and ethical AI.
'AI ethics' is a code of conduct that governs Artificial Intelligence, covering how it acts out in case of an ethical issue.
The list of 23 Asilomar AI Principles is an example. The principles are divided into three areas: research issues, ethics and values, and longer-term issues related to AI’s research, use, and more.
On the other hand, Ethical AI are moral principles governing the making and use of AI in your business. For businesses, this means having a framework in place that ensures that the AI you are building is ethical.
For example, when faced with the question of limits of ethical AI, Amazon promptly abandoned the use of its hiring algorithm that showed bias in favor of male candidates. The AI was trained using historic data, wherein male candidates were largely hired, so the algorithm followed suit.
Amazon’s example brings to light an important question: does your business have an ethical AI framework in place? Is the AI you’re employing showing confirmation bias or any other issue? And, how would you react in the case where the adverse happens as it did in Amazon?
This, in short, is ethical AI for business. Let’s learn more about how to set limits for ethical AI.
Why is ethical AI important?
Leading industry names like Bill Gates have already started voicing their concern about assessing risk before it’s too late.
Other than that, Eugene Goostman’s progress and Amazon’s faulty hiring algorithm set the need for an ethical framework for your business.
Here are more reasons why ethical AI is essential:
Errors in AI can threaten public security and safety
Often, errors can be unintentional. They can also be invisible to humans as to how Machine Learning works is still not fully clear.
Case in point: a fault in the AI of self-driving cars can jeopardize public safety. In case of an accident, an important question that remains unanswered is whether the robot should save a man on the street or someone in the car.
In the wrong hands, AI can be used to commit fraud
Not only can Machine Learning and AI help detect fraud, but they can also abet them. Hard to believe? Take this: in 2019, AI was involved in a scam of $243 million.
How? In the wrong hands, it was used to impersonate a CEO’s voice that succeeded in duping an employee into the hefty sum’s fraudulent wire transfer.
Unsuspected bias can lead to gender, race, and age biases
This threat closely relates to the example of Amazon’s AI that we talked about above. Under-trained algorithms or those that are trained with skewed data from the past can show confirmation bias.
Here’s another example: a UNESCO study confirmed that public-facing chatbots showcased potentially harmful gender stereotypes.
These are just some of the reasons why you need to pay attention to ethical AI business practices. Read on to learn about AI’s common ethical issues next.
What are the ethical dilemmas associated with AI?
From unemployment to trust deficit and bias problems, there are several ethical issues you need to be mindful of.
Let’s dive into these issues briefly. Then we'll address solutions for the most pressing ones.
1. Computing power
Artificial Intelligence runs on a vast number of GPUs and core that are costly. Similarly, implementing Deep Learning frameworks takes computing power from supercomputers. Again, these don’t come cheap.
There’s a third aspect to this: the ever-changing and complex nature of algorithms means that working on accommodating cloud computing adds to the cost.
All this makes the power and cost involved in the development and smooth running of Machine Learning and Deep Learning a cause for concern.
With Artificial Intelligence taking over jobs that employ humans, the subsequent unemployment that may arise is an ethical dilemma.
In fact, by 2022, AI is expected to eliminate 75 million jobs according to the World Economic Forum.
Take the trucking industry, for example. As of 2018, it employed over 1.5 million individuals in the US. However, with the use of self-driving trucks becoming widespread, how many of these jobs do you think will be lost?
3. Trust deficit
The lack of trust arises from deep learning models’ unknown nature as the human brain and algorithms work differently.
In other words, how AI works to the extent that hidden biases occur without human awareness is an ethical question that needs attention.
4. Data privacy and security
Training algorithms require large datasets. Procuring them is only part of the problem.
A serious ethical issue is a threat that the datasets post if they end up in the wrong hands and become subject to misuse.
5. Unequal wealth distribution
Several companies employ people on an hourly wage. With AI, however, these businesses can significantly reduce the human workforce, which would reduce the revenue going to them.
As for AI-driven businesses? They’d make most of the money. Of course, this widens the wealth gap.
6. The bias problem
Confirmation bias is a serious cause for concern for algorithms that aren’t trained by a diverse background or are fed biased datasets.
As a result, it’s common to come across Amazon-like algorithms that show bias in hiring and other work lines.
7. Questionable efficiency
In contrast with AI, humans are always reliable in terms of efficiency. For instance, software that sifts spam emails from useful ones will always have a 90% accuracy, where a human will show 99% accuracy.
For artificial intelligence to achieve such performance, you’d need a large dataset, unlimited fine-tuning, robust computing power, and uninterrupted training, among other things.
Not only all this requires deep pockets, but it’s easier said than done.
How to create transparency in AI decision-making
Now that you know why ethical AI for business is essential and the reasons behind setting limits of ethical AI, let’s discuss possible solutions for the most pressing issues.
Note that these suggestions come from the European Commission’s “Ethics Guidelines for Trustworthy AI.” The document enlists essential steps to take for developing ethical AI, and you’ll find some of the below:
1. Make sure human oversight is involved
The idea is to make sure that algorithms and Artificial Intelligence don’t run entire systems. Instead, human gatekeepers take charge of the decision-making process.
Alternatively, humans can be involved in the review process, if not the decision-making one.
Recently, Facebook let go of the human editors involved in its curated Facebook trending news section. The result? Not what you’d expect. The AI ended up publishing obscene and false material, and Facebook had to shut it down.
The take-home message? Humans reviewing the decisions or, as mentioned, taking decisions are crucial. This way, you won't find yourself in hot water.
2. Be clear on how statistics-based Machine Learning models work
Although you may not know exactly how Machine Learning works, putting the matter on the back burner won't help.
Instead, a viable solution for building an ethical AI framework for your business should involve taking proactive steps to understand how machine learning models work.
It's best be transparent about it as well.
3. Be sure to put in the effort to build AI systems that are resilient to adversarial attacks
As discussed, the way that algorithms work is inherently different from the human brain. And, it’s what sets the ground for adversarial attacks. These are small manipulations in the AI’s behavior that occur due to minute changes in the input data, invisible to humans in most cases.
Often, these adversarial attacks can be an accident. Other times, however, a malicious actor might be behind it.
A study, for example, learned that changing the appearance and coloring of a stop sign in an unnoticeable manner can confuse self-driving cars, causing a safety threat.
If you’re in the business of Deep Learning, you’re prone to such attacks. The solution, you ask? A good one is enabling your AI systems to fallback from Machine Learning to rule-based systems or ask for a human to intervene.
4. Observe transparency on data collection and processing practices
Several people object to the usage and storage of their data. They also question its access.
Therefore, having the answers openly available for your target audience is essential for building trust with them.
In 2019, for example, it emerged that a team of humans listens to your voice recording that the Echo device captures whether they’re purposefully activated or not. Not only does this fracture buyers’ trust, but the information led security experts to claim the smart speakers aren’t worth the purchase as they are “inherently insecure.”
Being transparent on data collection, therefore, is the ethical way to move forward. Some companies like Google, for instance, offer users the option to delete their data from their servers. Others allow users to download their data.
5. Make sure your AI systems’ datasets are inclusive to keep hidden biases at bay
Lastly, to prevent confirmation bias, essential ethical AI business practices involve:
- Solving the hidden biases by hiring people from diverse backgrounds, disciplines, and cultures to train the algorithm
- Preventing the risks of confirmation bias by ensuring embedded datasets fed into the systems aren’t skewed
These steps help make sure that the AI system can render more human-like results without showing biases.
Wrap up thoughts
Failing to operationalize AI ethics and data can lead to wasted resources, inefficiencies in product development, and even an inability to use data to train Machine Learning models at all, if biases are not corrected early on.
With no clear protocol in place companies end up overlooking risks or scrambling to solve issues as they come up.
Summing up, ethical AI for business is more important now than ever before. With transparency issues, confirmation bias, and the other risks that we’ve discussed here, it’s essential you set up an ethical AI framework for your organization while you're still at the start of your AI journey.