Stewarding AI responsibly: lessons from Affectiva's Rana el Kaliouby

Executive overview

Building transformative technology is not enough — founders must also actively govern how it is used. Rana el Kaliouby co-founded Affectiva to bring emotional intelligence to machines, but the harder work was drawing firm lines against surveillance, deception detection, and data exploitation, even when $40M was on the table.

The best stewards of an idea are those most deeply invested in its original purpose — and willing to say no to protect it.

Origins: from affective computing to Affectiva

• Rana read Rosalind Picard's book Affective Computing in 1998 and pivoted her entire career to build emotionally intelligent machines.
• Key insight from Picard: computers need emotional intelligence, not just cognitive IQ, to truly serve humans.
• Rana's PhD at Cambridge focused on software that used webcams to map facial expressions to emotional states.
• First breakthrough was a "head nod detector" — a proof that dynamic, temporal signals could be algorithmically captured.
• Mid-PhD realisation: the technology wasn't just about human-machine interaction; it was about human connection.
• Picard visited Cambridge, saw the demo, invited Rana to MIT Media Lab — a pivotal ally acquired by showing work.

Building the product at MIT

• The pair developed a Google Glass-style device to give real-time social cue feedback to individuals on the autism spectrum.
• NSF rejected the initial grant ("amazing idea, impossible to build"), so they built it first and reapplied.
• MIT's Member Week events surfaced 20+ industry sponsors wanting the technology — the tipping point toward commercialisation.
• Media Lab director Frank Moss told them: "This is not a research problem anymore. Spin out."

Founding Affectiva and setting stewardship values

• Before winning their first customer, Rana and Picard defined core values: consent, privacy, data reciprocity, and prohibited use cases.
• Explicitly ruled out: security, surveillance, lie detection.
• Every hire was self-selected around those values; culture became the enforcement mechanism.
• Picard: CEOs must be responsible for good use — not just release technology and let society sort it out.

The $40M test

• In 2011, with three to four months of runway, a venture arm of an intelligence agency offered $40M — contingent on pivoting to surveillance and deception detection.
• Rana visualised what the world would look like if they took the money. It conflicted with every core value.
• They turned it down without knowing whether they could raise elsewhere.
• Lesson: you will face decisions that could end your company. Protecting the vision means being willing to let that happen.

Finding product-market fit in ad tech

• Affectiva built an online panel asking users to watch ads with webcams on; thousands opted in.
• WPP approached as both partner and investor after seeing the crowdsourced emotion data.
• By 2015, ad testing was highly profitable and nearly fully automated.
• Rana recognised the drift: the mission was to humanise technology, not optimise ad campaigns.
• Course correction followed — not because they had failed, but because stewardship requires taking bearings.

Pivoting to automotive

• Inbound from Japanese and European luxury car brands drove a deliberate pivot to in-cabin sensing.
• Core ethics decision: all processing happens on-device; no video leaves the car; nothing is recorded.
• Live footage revealed alarming driver behaviour — people asleep, texting, teenagers drinking — even knowing cameras were present.
• Data validated the technology's road-safety potential and reinforced why the privacy constraints mattered.

Scaling stewardship beyond the company

• Affectiva joined the Partnership on AI consortium (including Amnesty International and the ACLU) to map all emotion AI use cases and risks.
• Founded the Emotion AI think tank: AI innovators, business practitioners, academics, ethicists — and competitors invited.
• Ongoing internal debate on surveillance: the team has revisited the question and still lands on no, citing data bias risks and lack of policy transparency.
• The argument for doing this work as a CEO: a broader definition of stakeholder — community matters as much as the next sales call.

The steward vs. the custodian

• Fine line between a thoughtful steward (nurturing the idea's best uses) and a cranky custodian (controlling how people use the product).
• Stewardship character is a mindset, not a CV credential: recognising that what you created is a public asset, not a personal property.
• Being judged on how the idea "plays in the world and in society" — not just on company metrics.
• Stewardship scales when others are invested: partners, consortia, think tanks, and even competitors share the burden of oversight.