6 Things Everyone Should Know About Artificial Intelligence [AI For Everyone Course Takeaways]

About 2 weeks ago, I started a course called AI For Everyone, taught by DeepLearning.AI founder and Coursera co-founder, Andrew Ng. This non-technical course is for anyone who is interested in applying AI to their business and learning about its impact on society.

As a UX designer and a techie, I love learning about breakthrough technologies. Artificial intelligence is definitely one of them. From smart speakers and self-driving cars, to resume scanning software that can decide your next job, AI is fundamentally changing the way we work and play.

Now that I’ve finished it, I hope to share 6 key ideas about AI which I got from the course.

Key takeaways

1. Understand the jargon: AI, ML, deep learning and data science
2. Most AI systems today are supervised learning systems
3. Choosing the right AI project is a team effort
4. A high level view of how to implement AI into your company
5. What a good AI company looks like
6. Like any world-changing tool, AI has its limitations

1. Understand the jargon: AI, ML, deep learning and data science

There are 4 key concepts that are taught in the course: artificial intelligence (AI), machine learning (ML), deep learning (DL) and data science (DS). AI is the overall topic; ML and deep learning are the techniques used in AI.

Today, when we talk about AI, we usually refer to Artificial Narrow Intelligence (ANI). ANI is a one-trick pony — it can do one specific thing very, very well. This is in comparison to Artificial General Intelligence (AGI), which is doing everything a human can do. We are still very far from the AGI, but ANI is changing the world today.

Unfortunately, we are still very far from Wall-E-like intelligence. Photo by Marius Haakestad on Unsplash.

Machine learning refers to getting a computer to produce a specific output without explicitly programming it to do so. In this case, it means that with input A, the computer will tell you output B. I’ll touch more on this later.

Deep learning is a sub-set of machine learning. It is used interchangeably with the term neural network. Basically, deep learning is made up of layers of mathematical formulae (each called a “neuron”) that build off each other, becoming more and more abstract. This allows a computer to “understand” complex ideas, and produce output B.

For example, a t-shirt seller might want to predict the demand the of a specific t-shirt. Given the price, material and marketing effort of a certain t-shirt design, the algorithm will first predict simpler concepts like affordability, awareness and perceived quality of the t-shirt. They then use these outputs to create another mathematical formula for predicting the more complex idea of demand for the t-shirt.

In this way, deep learning algorithms can appear to process very complex data, such as recognising faces, playing a song you asked for or even charting a path for a self-driving car.

Finally, data science is making sense of big data. You can think of AI as the machine that runs the factory, while data science is the pitch deck with all the business insights and recommendations.

AI is the entire topic. ML is a way to achieve AI. Deep learning is a specific form of ML. Data science intersects all these levels, and provides actionable business insights.

Knowing what these terms meant made learning about AI much less intimidating. If you’d want a more detailed explanation of these concepts, you can refer to week 1 of AI For Everyone course, where Andrew explains each concept clearly with plenty of relatable examples.

2. Most AI systems today are supervised learning systems

The most popular form of machine learning today is supervised learning. Remember when I talked about the input A and output B process earlier? This is what supervised learning is. Give A, get B.

Let’s illustrate this with an example. Say you are a manufacturer of coffee mugs, and you want your AI to detect if there is a defect in your coffee mug. You need 2 types of data to give your AI in order for it to “learn” what a defective coffee mug looks like. These are:

1. Many photos of normal coffee mugs (A), labelled as not defective (B).
2. Many photos of coffee mugs with scratches (A), labelled as defective (B).

You can get this data by having your employees take photos of your mugs, by collecting data with software, or by buying the data directly.

By giving it the photos (input A) and the desired label (output B), the software “learns” how to tell what a defective and normal coffee mug looks like. The next time you give them a photo of a coffee mug (input A), it should be able to tell you whether this mug is defective or not (output B).

3. Choosing the right AI project is a team effort

I learnt the main steps in transforming your business into one that is powered by AI. Most businesses have tasks that can be automated, saving time for employees to do more creative, non-repetitive work.

“AI is automation on steroids.”

- Andrew Ng, Founder of DeepLearning.AI

In the course, it is emphasised multiple times that AI is not a cure for everything; it cannot do everything. Thus it is important to do both business and technical diligence before choosing an AI project. Ideally, you’d want your first project to be both valuable and feasible.

Finding a valuable and feasible AI project takes both business domain and AI experts.

A good rule of thumb to determining technical feasibility is the one second rule. It goes like this: Anything a human can do in one second of thinking, AI can probably do today or very soon. To determine business value, you need deep knowledge of your industry, user problems and your unique selling points (USP).

This discussion on technical and business diligence is best done with business leaders and AI engineers in the same room. It’s where UX designers can help with brainstorming techniques and prioritisation matrices.

4. How to implement AI into your company

In the course, Andrew mentions the AI Transformation Playbook, a step-by-step guide on how to turn your business into an AI business. This 5-step plan is:

1. Start with small pilot projects. No project is too small. As a first project, you should select something that has a high chance of success and ideally can show results in 6–12 months. This will help build confidence and momentum for bigger projects in your company.
2. Build an in-house AI team. After showing some initial success, it is best to build an in-house AI team. This could look like hiring a few AI or ML engineers. However, it’s not compulsory. You and some colleagues can also take an AI course and gain skills to start the team.
3. Provide broad AI training. Next, you provide broad training to other executives, business unit heads and engineers throughout your company.
4. Develop an AI strategy. By now, you should have a clear idea where the biggest value AI can provide your company, so it is a good time to build an AI strategy. This will create a roadmap for how AI can benefit your business.
5. Communicate the strategy to stakeholders. Lastly, you communicate these changes to internal stakeholders, such as a board of directors, and external stakeholders, such as your customers.

The details of each step are too numerous to mention here. If you’d like to find out more, you can take a look at week 3 of the free AI For Everyone course, or download the AI Transformation Playbook.

5. What a good AI company looks like

I never thought I’d learn how to differentiate what good and bad AI companies look like so early in my career, but here we are. The main indicators of a good AI company are:

1. A unified data warehouse. This makes it easy for an AI or ML engineer to get the data needed to build their algorithms and software. If data is stored in siloes, it will be impossible for the engineers to operate efficiently and effectively.
2. Automation opportunities are easily acted upon. The best part of AI is automating repetitive tasks. A mature AI company will often automate these tasks as and when they spot them.
3. Strategic AI product development. AI requires a lot of data. That’s why some of the big tech companies create software that is not directly monetisable, but gather lots of data that can later be fed into an AI. For example, Google Maps can track shops you’ve been to recently, thus recommending you certain clothes or gadgets you could be interested in the form of ads.
4. Has AI and ML engineers. Usually a mature AI company has these specialised software engineers. However, you can undoubtedly start building an AI without these specialised roles!

6. Like any world-changing tool, AI has its limitations

AI is great at automating tasks and is already being applied to important fields like healthcare, security, consumer tech and self-driving vehicles. However, it is important to not be too optimistic nor too pessimistic about AI. This middle ground mindset is more accurate and is called the Goldilocks Rule.

We should have a balanced view of what AI can and cannot do. Photo by Jon Flobrant on Unsplash.

While AI is capable of much, it has its limitations. One example is that it can lead to biased outputs when fed biased data. One research paper found that when fed web articles from the internet, the AI produced the following output:

• Man is to Computer Engineer, as Woman is to Homemaker

This sexist stereotype is unhealthy, and should not be something that is promoted. Instead, the output should be:

• Man is to Computer Engineer, as Woman is to Computer Engineer, or
• Man is to Homemaker, as Woman is to Homemaker.

AI researchers are working hard to reduce the bias in AI through technical means, such as reducing certain biased words to the value 0. This is called “zeroing out” the biased words. Another way to reduce bias is to simply feed the algorithm unbiased data from the start.

There are many more impacts of AI on society, including but not limited to its impact on race, jobs, security and developing economies. If you’d like to find out more, you can take a look at week 4 of the AI For Everyone course.

Conclusion

AI is changing the world, one piece at a time. As the technology is still so new, many individuals and nations alike have the time to grasp this opportunity. In the future, AI will touch multiple aspects of our lives, whether or not you work in the tech industry. That’s why I believe it’s important to learn about it whether you’re a designer, or just a fellow human living on this planet.

Hi! I’m Rebecca, a UI/UX Designer based in sunny Singapore. Having used design thinking as both an in-house designer and freelancer, I have worked on both smaller, usability issues and long-term, complex problems. In my spare time, I can be found passionately reading up on sustainable design or playing squash (the sport, not the vegetable). Connect with me on LinkedIn, Instagram or my website.