Today we release Mission Control; the project management platform for better data science.
We invite qualified teams to request access to our Discovery Program. Learn how Mission Control helps the world’s leading companies move data science faster and break fewer things at www.takecontrol.ai.
I’m also honored to share news of our recent $2M funding round, led by Stage Venture Partners with significant participation from OCA Ventures, Royal St. Ventures, Protocol Labs, and Vibe Capital. We’re extremely grateful for their vision and conviction. We could not be more enthused to welcome these brilliant men and women to our team.
We’re AIRL
We’re AIRL
We’re The AI Responsibility Lab.
A Los Angeles-based Public Benefit Corporation focused on AI Safety.
We build venture-scale software that makes it easier for companies to use AI safely from the start.
That starts with Mission Control: the project management platform for better data science. Mission Control empowers high-velocity data science teams to detect problems before they happen, transform notebooks into scalable playbooks, and get real about Responsible AI.
While researching the growing world of AI Safety and Responsible AI, we discovered two very important things that differentiate Mission Control.
1) AI Ethics isn’t the problem. It is the symptom.
Other tools in our space talk about “Ethical AI” and “Responsible AI”.
These terms are well intentioned. They point to a foundational concept(ethics; the underpinning our understandings of The Good Life), and an end-state (something like the futuristic version of Corporate Social Responsibility: Corporate AI Responsibility). We can’t expect teams use data better without knowing fundamentals (ethics) and a goal (responsibility).
But few data scientists think about their problems through these lenses. The incentives for ethics aren’t strong. They’re incentivized for project velocity and model accuracy.
2) The root problem is Coordination Failure
Coordination Failures are what happens when independent actors must act cooperatively towards a mutually beneficial common outcome, but nevertheless fail to do so.
These failures emerge [at least] from a variety of reasons; information asymmetry, high communication delays, [perverse] incentive structures, organizational opacity, system complexity, and cognitive biases.
For example:
“The credit worthiness prediction algorithm produces sexist outcomes”.
That’s a symptom.
“We aren’t checking training data for fairness before training a model on it. The Head of Data Science didn’t know we weren’t doing this. We don’t know who should be in charge of this. We don’t know how they should report on it, and to whom they should report. Honestly, we don’t totally know where this dataset came from. Brent knew, but he left. We don’t know how many other models have this problem now, too. That was a Brent thing.”
That’s a root cause. That’s a coordination failure.
They are the root causes of the types of problems that become Responsible AI problems. When we realized this, we know what we had to build.
If you want more ethical outcomes for AI, make it harder to have coordination failures.
If you want fewer coordination failures, make tools that make it easier for leaders and practitioners to coordinate.
If you want teams to make sense of it, connect it to their incentives; better work produced faster with fewer defects.
So we built Mission Control.
Mission Control
Mission Control
Mission Control is the project management platform for better data science.
It’s a web platform for leaders to manage their data science team. It marries project management best practices with the unique needs of data science teams.
It empowers teams to detect and fix problems before they happen. To transform one-off notebooks into scalable, repeatable workflow playbooks. To uncover project complexity and predict success. To manage systems and people like a team. To make realistic steps towards improving the safety, stability, and speed of their work.
Mission Control is a dashboard and a series of interconnected modules that improve workstream visibility, team coordination, project consistency, business continuity, ROI, and project velocity.
Problem: We can’t catch avoidable mistakes early
in a data science project’s maturity.
Problem: We can’t catch avoidable mistakes early
in a data science project’s maturity.
Many teams detect problems with their datasets, policies, and models once it’s too late. Workstream Intelligence automates the detection of potentially problematic data science scenarios and empowers your team to take back control.
By automatically detecting high-risk data science situations — like fairness and explainability violations — before they become problematic, Workstream Intelligence monitors and automatically flags potential issues. This helps you catch mistakes earlier in a project’s lifecycle and improve your ROI and deployment velocity.
Problem: Our data science work is inconsistent. This is starting to hurt as we grow.
Problem: Our data science work is inconsistent. This is starting to hurt as we grow.
A team of 2 can do ad-hoc data science. Not a team of 10. Especially when you’re remote. Worse; as your team grows, your median years of experience will drop. More people, less intuition.
Workflows are customizable, flexible, full-lifecycle project rails. They help you standardize and scale best practices as flexible project work templates. They allow you to transform notebooks into repeatable playbooks. This empowers you to scale your best practices and help your team do better work faster with clear expectations and support. This helps you do more consistent, zero-defect work faster with less managerial overhead.
Problem: We don’t know how our moving parts connect. We can’t predict success or failure.
Problem: We don’t know how our moving parts connect. We can’t predict success or failure.
When teams grow, they lose track of how the pieces fit. This leads to unwanted surprises; usually in the form of defective projects, scope creep, low velocity, and problems caught too late. Complexity Navigator helps your team drill down into the network of connections that power your success.
The complexity of a data science project isn’t the number of moving parts; it’s the number of relationships between the parts. Complexity Navigator helps you visualize relationships; between people, datasets, models, workflows, and policies — and make better decisions faster.
Problem: On Monday, Jen’s team trained Model A on Dataset B. On Tuesday, Dan’s team reviewed and rejected Dataset B for use. Jen and Dan don’t know one another exist.
Problem: On Monday, Jen’s team trained Model A on Dataset B. On Tuesday, Dan’s team reviewed and rejected Dataset B for use. Jen and Dan don’t know one another exist.
Many cross-functional teams will contribute to a single project before it hits production. They don’t always communicate well. They don’t always know one another exists. Dataset and Model Management helps you register your models and datasets against one another, and the teams, workflows, and policies they’re bound to.
Model and Dataset Management makes it easier to understand who’s working on what and ingest data streams from monitoring systems.
Problem: We keep hearing “Responsible AI”. No idea what that *actually means.
Problem: We keep hearing “Responsible AI”. No idea what that *actually means.
Responsible AI is trending. But how do ideals and ethics translate into how we run a data science team?
Gap Analysis and Gap Prevention turn abstract concepts into practical changes your team can make. They were designed by globally-recognized Responsible AI experts to help you do better work, with less work.
No one intends for unethical outcomes from their data systems. But they happen. Gap Analysis and Prevention help you spot their root causes earlier and take small, templated steps to fix them, designed by globally-respected Responsible AI experts.
Our Roadmap
Our Roadmap
We’re an AI Safety company. But what does that mean?
Broadly, AI Safety is the family of work that seeks to maximize the upside from AI, while minimizing the downside and the likelihood of terminal failures.
It’s equal parts philosophy and engineering. Cognitive science and public policy. Its focus spans different timescales; from near-term AI Ethics or Responsible AI, to long-term Existential Risk from malevolent superintelligence.
We break the timeline down into 3 chapters; 2022, 2027, and 2035.
AIRL is actively developing solutions for all 3 chapters.
- For 2022, we release Mission Control. Mission Control aligns near-term challenges and incentives to make it easier to scale safe AI from the start.
- For 2027, we release SLIP: the Synthetic Labor Incentive Protocol. SLIP incentivizes autonomous AI economic actors to comply with manmade AI Safety protocols.
- For 2035, we release Hofstadter Detectors. Hofstadter Detectors probe AGI for signs of consciousness, starting with self-referential behaviors and thought patterns.
AI Safety takes everyone. Workers in the field have a lot of different roles. Some working groups write papers. Other build new ways to make existing AI safer. Some make entirely new, safer AI. Some focus heavily on social justice implications. Others on the environment. Some build narrow tools that make one facet of AI safer. Others (like us) create platforms that make it generally safer to build new AI.
All in the service of building a better, safer world of AI. And with not a moment to lose.
We have Two Paths forward:
We have Two Paths forward:
AI is the most important invention in human history.
It’s 2022. Enterprise adoption of AI is happening at the same time that corporate labs are releasing v1.0 of Artificial General Intelligence (AGI). The world will soon be full of generalist AI agents capable of robust skills, transferring learnings from one task to another, and performing well under new circumstances with little or zero retraining.
We are very close to synthetic thought. Man-made intellect. Manufactured souls. 1 trillion autonomous minds on the net. A pace of technical and social change measured in seconds, not years.
And to the best that we can tell, There are two paths forward with AI:
The first path is towards unprecedented flourishing.
In Path #1, the incentives, ownership, and returns of AI are to the net benefit of all of humanity. Humans and synthetic intelligence work and coexist side-by-side as partners. Solutions are found to problems we otherwise viewed as intractable. Unprecedentedly high productivity is combined with unprecedentedly low labor participation. We have enough for everyone.
The second path is toward unprecedented languishing.
Today, AI already exacerbates existing structural violence around sex and race. Systems control critical decisions in our lives — who lives and dies, who is granted economic inclusion — and behave capriciously. Ethnic cleansings have been exacerbated by algorithmic optimization. We’re in for another jobless economic recovery in 2024 as jobs automate in 2022–2023 in response to market pressure. There’s an oft-hidden, massive ecological impact in raw materials and energy that goes into AI. Critical infrastructure is increasingly unstable, unusable, and vulnerable to disruption and attack. We’re losing fundamental rights to privacy, the sanctity of our personal data, and autonomy itself.
All of these harms are real, today. All of them hurt all of us. They hurt the companies building the AI systems that create these outcomes.
Everyone — society and the private sector alike — is strongly incentivized to fix this. No one wants Path #2. It’s not dignified. It’s not graceful. It’s not profitable.
So that’s why we’re accelerating AI Safety. To get us on Path #1.
We’re grateful to our families, friends, partners, communities, and investors for their faith, patience, support, enthusiasm, conviction and love in helping us get here. We can’t do this without them. Thank you all. We love you.
Join us
Join us
If this mission speaks to you and you want to get involved — as a team member, expert, investor, or partner; reach out at hello@takecontrol.ai.
If you’re an AI Safety leader or practitioner and you’d like to join The AI Safety Slack, apply here: https://airtable.com/shr2dXkZJIWmvF4sV.
~
-Ramsay Brown. CEO, Founder @ The AI Responsibility Lab PBC. May 23rd 2022. Los Angeles, CA.