Artificial Intelligence | Lesson 1.7
What AI Can and Can’t Do
This Nonprofit Uses Artificial Intelligence To Non-Invasively Monitor Endangered Species by Business Insider
We hope you develop a foundational understanding about what AI can and cannot do in this section. In practice, before we commit to a specific AI project, it is usual to have either ourselves or engineers do technical diligence on the project to ensure it is feasible. This means looking at the data, the input and output A and B, and just thinking through if this is something AI can do.
One of the challenges is that the media and academic literature tend to only report on positive results or success stories using AI. People sometimes think AI can do everything when we see a string of success stories and no failure stories. Unfortunately, that is just not true.
- AI lacks emotional intelligence and empathy
- AI can only work with inputted data
- AI does not have soft skills
- AI’s creative process is limited to the data it receives
- AI needs to be fact-checked
- AI fails in capturing or responding to intangible human factors such as the ethical, moral, and other human considerations that guide decision-making.
- AI isn’t ready to make unsupervised decisions.
It would help if you saw a few examples of what today’s AI technology can do, but also what it cannot do, and where things have gone south. Some examples where AI’s decision-making capabilities went wrong are highlighted in a 2022 article from Harvard Business Review, which includes Microsoft’s AI teen chatbot TAY (stands for Thinking About You). The chatbot was built to engage with millennials in a casual way and was designed to learn from interactions with real humans. TAY learned offensive language and incorrect facts from other users in the learning process, and didn’t engage in fact checking. Within 24 hours of launch Microsoft killed the bot. CNET reported the following in 2016: ” A coordinated attack by a subset of people exploited a vulnerability in Tay.” Peter Lee, a corporate vice president of Microsoft Research, stated. ” As a result, Tay tweeted wildly inappropriate and reprehensible words and images.”
To help you hone into what might be more or less promising AI projects to select for your organization, a good rule of thumb is to break the task into two different categories inputs and outputs as we mentioned before, to determine what AI can do. We described what inputs and outputs are earlier. You need to think ahead in case the output is not available. Can an individual give an estimation of the output based on the inputs you have? If they can, AI is a great method to automate that process.
What inputs do you assume you always have access to, and outputs? What do you want to estimate? So, implementing AI is possible if you can break your task down into inputs. For example, in the house price prediction app, the task can be broken down into house features (inputs) and the price (output).
The real estate brokers use the same set of inputs every day to give people price estimation. Hence, it is a great choice for AI. This is a great starting point. However, it is not sufficient. In the next sections, we discuss in more detail how to develop an AI project and ensure it is feasible.