EchoVC Partners

View Original

Our Investment in KOSA

EchoVC is excited to announce its investment in KOSA.ai - an artificial intelligence startup that offers an automated responsible AI platform to support enterprises that build and deploy AI systems.

Artificial Intelligence (AI) will affect consumers and enterprises unlike any other technology we have seen before and has been described as the greatest economic opportunity of our lifetime - estimated to contribute roughly $16 trillion to the global economy through 2030 ($8.7 trillion excluding China) (1).

For enterprises, AI delivers automation and valuable insights that enable companies to scale more efficiently, and provide better services and products.

In 2020, a survey by IBM found that ~75% of enterprises across the United States, Europe, and China had either deployed AI or were ramping up their exploratory AI phases. IBM estimates this percentage will jump to 80-90% within the next 18 to 24 months (2).

However, a major headwind to enterprise AI adoption is the limited in-house expertise to build and develop AI models, and manage data complexities.

One consequence of this is AI bias - affecting large and small enterprises alike.

AI bias is the manifestation of systematic prejudice and unfairness in the results of artificial intelligence algorithms. This can arise from incorrect training or real-world data that is used to develop the AI; or from the conscious or subconscious bias of developers creating the AI. Such bias may not even be apparent until post-development once the AI model begins to "drift" when deployed into the real world.

Bias in AI creates error rates, failed transactions, as well as missed or under-served consumer pools - all of which result in lost revenue for the enterprise. Other consequences include a weaker value proposition to the enterprise's customers, lower competitive ability, as well as heightened regulatory risk and brand reputation risk.

Some past occurrences of AI bias have included:

Amazon

- In 2019, Amazon's facial recognition AI, Rekognition, was found to have bias-related errors in identifying women and darker-skinned individuals: it mistook women for men 19% of the time and mistook darker-skinned women for men 31% of the time. (3)

- Separately in 2015, Amazon realized that their hiring algorithm had incurred bias against women since it had been trained on résumés submitted over the prior 10 years - which were mostly male résumés. (4)

Facebook

- Was sued in 2019 by the United States government because its ad-targeting algorithms had developed gender and ethnic bias in how it chose which consumers to target for housing and job ads. (5) 

Self-Driving Algorithms

- Are also prone to bias-related errors, with a study finding that Black pedestrians, compared to white pedestrians, were 5% more likely to be hit by self-driving cars due to biased object detection models. (6)

UnitedHealth Group:

- A study published in Science in 2019 found that UnitedHealth’s Optum algorithm, which services over 200 million people and predicts which patients would likely need extra medical care, heavily favored white patients over Black patients - denying care to nearly 50% of Black patients in need (7)

Apart from lost revenue, lower competitive ability, and brand reputation risk, AI bias will soon expose enterprises to significant regulatory risk.

Earlier this year, the European Commission released legislative proposals on AI governance with contemplated fines of up to 4% of global annual revenue or €20m (whichever is higher), for non-compliance of AI use-cases.(8) 

In the United States, a bill was introduced in 2019 to the US Congress, titled "The Algorithmic Accountability Act", which sought to mandate companies to “conduct impact assessments and reasonably address in a timely manner any identified biases or security issues” in their AI algorithms.

We believe that bias in AI is now, and will become a materially more, crucial problem for enterprises, and society, in the future.

KOSA can help solve this. 

KOSA is a data analytics and algorithmic company that assists enterprises to detect, audit, and explain bias in their AI models - and then implement corrective steps to address or mitigate the bias. Additionally, KOSA can support the enterprise in monitoring their AI models post-deployment for any "drift" towards bias. Furthermore, KOSA enables enterprises to define both AI KPIs and "fairness" for their AI models.

We view KOSA as a multi-vertical, layered approach to participating in the AI sector. Our investment thesis is centered on backing two exceptional founders, tackling a difficult - yet inevitable - problem; in a massive opportunity set. 

Layla Li and Sonali Sanghrajka, the co-founders of KOSA, impressed us with their backgrounds, their agility and ambition; and their vision to scale KOSA into a global business - potentially becoming an industry-standard in "fair AI".

We are grateful to partner with KOSA on this exciting journey. As part of this pre-Seed round, and in conjunction with APX, EchoVC was joined by members of an EchoVC-led syndicate comprising Dale Mathias, TheContinent Venture Partners, Fine Day Ventures, and Arch Capital.


See this content in the original post

(1) PWC: https://www.pwc.com/gx/en/issues/data-and-analytics/publications/artificial-intelligence-study.html

(2) IBM & Morning Consult: From Roadblock to Scale. The Global Sprint Towards AI. 2020.

(3) https://www.theverge.com/2019/1/25/18197137/amazon-rekognition-facial-recognition-bias-race-gender

(4) https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G

(5) https://www.wsj.com/articles/facebook-shows-men-and-women-different-job-ads-study-finds-11617969600

(6) https://scs.gatech.edu/news/620309/research-reveals-possibly-fatal-consequences-algorithmic-bias

(7) https://science.sciencemag.org/content/366/6464/447

(8) https://www.loc.gov/item/global-legal-monitor/2021-05-26/european-union-commission-publishes-proposal-to-regulate-artificial-intelligence/