AI for Good: Understanding AI and its Applications

By Foteini Agrafioti and John Stackhouse
Published July 5, 2019 | 4 min read

As artificial intelligence ramps up toward early commercialization, AI For Good is top of mind. But what are the opportunities, risks, and timelines associated with this world-changing series of new technologies?

Foteini Agrafioti, Head of Borealis AI and Chief Scientist Officer at RBC, sat down with John Stackhouse, Senior Vice-President, Office of the CEO, to bring clarity to what AI looks like today, to sift through the myths around what AI isn’t, and to discuss AI’s massive potential through responsible and safe applications of these world-changing technologies.

7 Things You Should Know Now 

1. We’re already using AI and Machine Learning every day 

  • There’s still a commonly held belief that AI is “robots” but the truth looks a lot more like algebra than Optimus Prime. At present, what we call artificial intelligence is actually a series of algorithms that enable computers to detect patterns at rates that surpass human expertise, then use those patterns to predict better outcomes.
  • These algorithms rely on data in order to learn and improve, so anywhere you’re already using real-world data and trying to make sense of it, you’re likely interacting with AI.
  • Some examples of what AI looks like in our everyday lives include the movie recommendations that pop up on Netflix each time we log in, or Amazon suggesting products based on your past purchase and search behaviour.
  • As these technologies improve, we can make huge strides in areas like healthcare diagnostics, infrastructure and translation services that will vastly improve quality of life at low-cost and on massive scale.
 

2. AI today is not what it can be tomorrow 

  • While our progress over the past few years has been staggering, AI has still not advanced to the level where it can safely make autonomous decisions for us.
  • We also don’t yet understand the full workings of the human brain during high-level reasoning in a way to represent that same intelligence in machines.
  • This is a very good thing. Until we understand why algorithmic models arrive at their decisions and what steps we need to get there, humans need to remain in the loop.
  • AI’s future potential will accelerate once we reach better explainability and can make the right choices on how to bring AI products to market.
 

3. AI is having a PR crisis

  • AI has been getting some bad press that is mostly rooted in poor visibility and understanding of what AI is – and perhaps more importantly, what it isn’t.
  • This fear began over the threat of mass-scale job loss due to automation, which has in too many cases been reported without proper context about new job creation and economy adaptations.
  • Bias – and the missteps that can arise from biased data – has further forced the conversation into action around fairness, transparency and accountability.
  • There is serious work being done right now around identifying potential misuses of AI, and how to build safeguards into algorithms in order to maximize AI’s potential for good. This work is coming out of academia and industrial research partnerships, as well as informed policy and legislation at the government level.
 

4. Be aware of the Big 9

  • The narrative around certain countries dominating AI would be better understood as select US and Chinese companies holding the power and major concentration of AI talent today.
  • These companies, coined “The Big Nine” by futurist Amy Webb, also hold majority market share and thus make it virtually impossible for anyone to compete at scale.
  • The question, however, is not how they use AI, it’s all the opportunities that are missed because this small pool of dominating players directs the course based on their own needs (advertising, entertainment, gamification, etc.) instead of areas at a critical juncture point that require real resources to solve now (climate, transportation healthcare).
 

5. AI is not straightforward… and may never be 

  • You can’t just plug-and-play AI and use it for your business anyway you see fit; there are always research steps that needs to be involved and that research requires highly specialized expertise.
  • For example, if you are an oil company and you want to figure out a carbon-efficient way to produce oil, you can’t simply buy an algorithm-in-a-box and apply it to your business. You need proper AI researchers to apply the specific data sets of your business to the problem in order to develop a custom algorithm based each unique business or application.
  • There is no general algorithm that can be applied to a problem. Every algorithm is unique and should only be used as a guideline.
 

6. Clean data and early detection are key to battling bias in AI

On Clean Data

  • We need to go back to the fundamentals to ensure our data is free of bias, that the data was collected in a transparent way, and is inclusive of the entire population.

Detecting Bias

  • It is important to build systems that act as a secondary defense on the main AI system.
  • Secondary systems like this would be tasked with monitoring the primary system and would be used to determine both how likely the system is to make decision ‘A’ or ‘B,’ and where there is a pattern detected.
  • This system of checks-and-balances will allow us to be fully transparent, identify if a system has built-in bias, and in case it does, it can react accordingly.
 

7. AI has the power to deliver profound transformation to our world 

Here are some examples of how AI can make the world a better place:

Autonomous Driving:

  • The ability for self-driving and self-powered vehicles could dramatically reduce congestion, accidents, and traffic circulation.
  • Humans are independent intelligent agent on the roads, each acting completely autonomously without the same logic in the background.
  • Autonomous cars would replace this with an intelligence that makes consistent decisions in every single vehicle.

Healthcare Sector:

  • If you are diagnosed with cancer today, you are given the same treatment as anyone else in the world diagnosed with the same form of cancer. But each person is different and reacts differently to treatment, meaning drug reaction is unpredictable and isn’t taken into consideration
  • With AI, this process would be personalized, identifying the right drug for the right person based on a number of measurements about each patient.
 

Listen to the 32 minute podcast now


Foteini Agrafioti and John Stackhouse


AIArtificial IntelligenceAutonomousDeep LearningMachine LearningRBC Disruptors