What can I make with AI right now?
A continuously growing collection of videos demonstrating what AI is good at solving for users today
Introduction
A bit about why we are where we are with AI today.
Predicting, automating, and optimizing time. That’s why you want to do AI.
Rob Thomas, GM for IBM Analytics
For the past 50+ years, AI’s progress was mainly limited by our inability to create faster computers, as well as coming up with a natural way for us to interact with them. Now that we’ve solved those issues well enough for the time being, we’re bumping into two new, but equally daunting challenges that are preventing us from fulfilling our sci-fi fantasies.
First, we can’t get data fast enough because, hey, who knew that’s what we should’ve been working on rather than widgets and album carousels?
The growth rate of worldwide big data revenue is estimated to jump from $49 billion in 2019 to $103 billion by 2027.
International Data Corporation, 2018
There are two solves for anyone facing a data deficit:
- If historic data is necessary for the accuracy of your model (ex. your model maps weather trends or traffic patterns), it’s going to take you some time to gather and digitize old data files, or to capture, label, and train new ones.
- If historic data isn’t a factor, or simply doesn’t exist (ex. your model is about making quality recommendations or diagnosing a new disease), then focus on finding new ways to train models using lower quantity, but higher quality, data. This usually requires subject matter experts and data creation services like those provided by Figure 8 and Mechanical Turk to author the data.
“If you look at all the applications where deep learning is successful, you’ll see they’re in domains where we can acquire a lot of data. The problem is not really about finding ways to distribute data, but about making our deep learning systems more efficient and able to work with less data.”
Professor Neil Lawrence in “These are three of the biggest problems facing today’s AI”, The Verge, Oct. 2016
Coming in at a close second is the lack of AI expertise in the world. First and foremost we need data scientists, but also AI focused domain, legal, ethical, operational, design, management, support, and training maintenance experts.
Demand for data scientists is off the charts. Shortages are present in almost every large U.S. city. Nationally, we have a shortage of 151,717 people with data science skills, with particularly acute shortages in NYC, SF, and LA.
LinkedIn Workforce Report, August 2018
Given these two speed bumps, where you see AI growing the most is in the machine learning sector. Machine learning is tool that can actually train AI to solve problems for us–hence, everyone is hyper-focused on writing an awful lot of algorithms these days–even if the data fueling these models isn’t ready yet.
If you combine our lack of data,
+ our lack AI experts,
+ our focus on ML,
then what AI can do best today tends to fall into six categories:
Automation
Conversational interaction
Insights, predictions, recommendations, and support
Optimizing efficiencies
Personalization
Transparency and training
Think of these examples like the AI version of simple machines. Right now we’re focused on getting the simple pieces built, then we’ll be able to combine them to create compound AI machines, like worker-less factories and AI powered nursing homes.
We’re starting to see attempts at compound AI use cases already that get more wow factor by stringing several types of ML together — like analytics + NLU + sentiment analysis — then combining them with a flashy delivery system like voice recognition, IoT sensors, or AR visualizations.
Just like determining what AI capabilities are growing the fastest, we can use the same approach to figure out which domains are using AI the most — it’s the industries that have the most money that are putting it to use. The big players currently include:
- Banking, financial services and insurance
- Legal and policy compliance
- Sales and marketing
- Manufacturing
- Retail
- Healthcare and life sciences
- Telecommunication
- Energy and utilities
- Government and defense
That said, regardless of how you want to use it or what industry you’re in, if you’re trying to determine whether AI might be a good tool for solving the problem you’re working on, an easy way to find out is to ask yourself…
Could it provide my users
better insights,
with more confidence,
faster than humanly possible?
If you can answer yes to one or more of these points, then it’s worth investigating more closely how AI could benefit your users.
Now let’s take look at real world examples of today’s six AI capabilities.
If you have case studies or capabilities you think should be added, please leave them in the comments section below.
AI Capability 1: Automation
Any small or repetitive tasks your user or business performs should be considered for automation.
AI Capability 2: Understanding text and answering questions in natural language
There are many moments where AI can bring relevant recommendations, predictions, insights or support to users when then need it in a conversational way that doesn’t distract them from what they’re doing.
AI Capability 3: Analyzing data for patterns and anomalies, then deriving insights and recommendations
If your product helps users to manage an organization, a system, or a team, then it’s likely that AI can help streamline the analysis process — even extend it into the future.
AI Capability 4: Understanding needs and recommending solutions
If your product analyzes any kind of data, then it’s in a position to bring insights, predictions and recommendations to your users. You can take it one step further by providing support when they appear to be struggling, have a question, or need to be guided through learning something new.
AI Capability 5: Personalization
If your product collects data about users behaviors, settings, or system, then AI can help improve their experiences more and more over time.
AI Capability 6: Providing explainability and allowing users to directly or indirectly improve future responses
If your product provides recommendations, insights, or predictions, it should always allow for progressive disclosure options into how it came to those conclusions. Also, don’t forget that your users are also subject matter experts. Give them every opportunity to tell you how to improve the experience, or even send their feedback directly to the model so it can retrain itself.
Thanks to https://appliedai.com for their use case reference library which I drew on extensively for this article.
Jennifer Sukis is a Design Principal for AI and Machine Learning at IBM, based in Austin, TX. The above article is personal and does not necessarily represent IBM’s positions, strategies or opinions.