In the iridescent flame that is 2020, you log online and see everything that is fueling the fire. Among those things is the controversy surrounding bias algorithms.
Throughout this year, you have been bombarded with information from all angles about topics of morality and how to make decisions based on the greater good. If you looked closely this topic of “ethical AI” might have popped up on your news feeds.
Unpacking the Term “Ethical AI”
It’s kind of like an angel-and-demon concept. If angels do exist, then surely demons have to exist.
The same goes for AI. If bias algorithms exist as the darkest side of AI, then there must be a shining light that can defeat bias data collection and ML engineering. There must be something to repair power imbalances, restore justice in facial recognition technology, and a guide to get ML outputs that won’t negatively affect someone’s life.
So it would be easy to say, ethical AI is unbiased data collection and engineering.
It’s Not That Simple…
In 2016, 74 sets of AI principles were published by the Berkman Klein Center at Harvard University.
The report collected up to 80 data points and represented them in the data visualization above.
The report reads,
A large variety of actors are represented, from individual tech companies’ guidelines for their own implementation of AI technology, to multi-stakeholder coalitions, to publications from national governments that incorporate ethical principles as part of an overall AI strategy. We expected to find some key themes, and indeed we uncovered eight: accountability, fairness and non-discrimination, human control of technology, privacy, professional responsibility, promotion of human values, safety and security, and transparency and explainability.
This report was known as the “first wave” of ethical AI. It was as if this was a reaction to the speeding up of algorithms in our every day life. The political events of 2016 perhaps lit the fire for the tech industry to start setting rules and standards for algorithms that could possibly have history-altering consequences.
In a recent article by VentureBeat, “This first wave of the movement focused on ethics over law, neglected questions related to systemic injustice and control of infrastructures.”
The article goes on to discuss the second wave of AI ethics saying it “narrowed in on these questions of bias and fairness, and explored technical interventions to solve them.”
In 2018, we saw some tech giants address bias and even introduce tools and ethics boards to identify bias:
Again, this new “second wave” of AI ethics was criticized as narrowing the focus to technical aspects, covering up the real issues at hand, and disregarding systematic injustices as we move forward in this new AI-driven world.
In a 2019 article Karen Hao said,
But talk is just that — it’s not enough. For all the lip service paid to these issues, many organizations’ AI ethics guidelines remain vague and hard to implement. Few companies can show tangible changes to the way AI products and services get evaluated and approved. We’re falling into a trap of ethics-washing, where genuine action gets replaced by superficial promises.
Should we define ethical AI as an overarching concept of an unachievable goal?
Before We Define Ethical AI, Let’s Discuss How We Move Forward
Now we’re deep into 2020. Facial recognition, deep fakes, and privacy concerns are at the forefront of AI. These topics may dominate the conversation as AI gets democratized. Even though it may not seem like it, 2020 was a huge step in AI accountability.
Several cities kicked out facial recognition use by law enforcement. Amazon, Microsoft, and IBM stopped selling their facial recognition tech. Google and Microsoft (among others) have been increasingly vocal against using AI to track migrants. Apple was called out for being biased in their credit card spending limits. A proposed bill could end facial recognition in public housing. How we view AI has changed significantly. We now have a better idea of the dangers of it.
Here are the key thing we identified.
Before a Company Deploys AI, They Should Build an AI System Correctly
Our last article by Eshita Nandini, discusses what companies using algorithms need to take into account when deploying models. A big theme was accountability.
The main challenge for for startups has become how to thoughtfully and emphatically build these machines — and less about if it is possible to.
While this piggybacks off the second wave of ethical AI, we must look into the technical aspects of building unbiased models while also addressing systemic AI pitfalls.
Read Eshita’s full article here.
The Current AI Industry Needs a Makeover
With stats like “Women are only represented in 18% of conference publications” and “71% of all job applicants for AI jobs in the US identified as male” it’s easy to see why the problem goes beyond data.
When we say “AI needs a makeover” we mean AI should be democratized and made transparent. Teams should be diversified and AI should be able to be communicated in non-technical terms to get over the large barrier to entry facing minorities today.
At Facebook, 100% of the AI engineers have a bachelor’s degree and 60% have a masters degree.
Big tech companies want their engineers to be proficient in:
- Java: 60%
- Artificial Intelligence: 57%
- Software Development: 48%
- C++: 38%
- Linux: 37%
- Python: 36%
While education is important in building algorithms, we must make it easier for AI skills to be obtained through ubiquitous education programs, transparency, and inclusive communities to share knowledge.
Read more about the AI industry here.
Let’s Listen to the Leading Voices in AI Ethics Instead of Large Companies
I highly encourage you to follow what leading voice are saying in the realm of AI ethics. At Obviously AI, we're stans for Joy Buolamwini, Timnit Gebru, Deb Raji, Devin Guillory, Nando de Freitas, and those leading the fight for fair ML.
Did We Even Define What Ethical AI Means in 2020?
To conclude, the term “Ethical AI” goes beyond unbiased data collection and algorithm accountability. Ethical AI is a movement towards a better AI-driven world to fully harness the power of AI systems. Ethical AI is always changing. While we are arguably on the “third wave” of ethical AI, there is still many discussions to be had on what ethical AI truly looks like.
In a Twitter thread, Joy Buolamwini argues her non-profit, the Algorithmic Justice League, uses poetry and storytelling instead of just research and data as a substitute for systemic change. This more contextual approach might be more common in the future of ML fairness as we also build a more inclusive and humanizing AI industry.
What the definition of ethical AI means looking ahead, I guess we just have to wait and see.