AI agents trusted to intervene in critical workflows

// Do more, faster and safer with our causal AI agents and human-machine teaming suite

Tulpa. Deploy Artificial Intelligence software that understands ‘why’.

// For considered, reasoned decision-making, human wisdom is irreplaceable. That’s why, for high-stakes, time-critical, exponentially complex challenges, expert pre-trained AI can help.

Generative AI and reinforcement learning are powerful techniques for creating AI agents, but they can be unreliable or difficult to interpret, posing risks and raising safety concerns. Our approach to developing and deploying agents is grounded in the need to ensure that a human can interpret and control what those agents do, and to understand why. This high-trust relationship between a human and machine intelligence is what we call ‘human-machine teaming’.

Combining behavioural science and machine learning, we rapidly capture and encode knowledge from experts. Our agents are able to fully explain why they took, or are proposing to take, a course of actions, which is a significant boost to assist and inform the work of experts.

For high stakes, mission-critical decision-making, bring human and artificial intelligence together with Tulpa.

Machine-augmented insight. Faster, considered decision-making with causal AI.

// As humans, we make decisions using both data and context. Our brain factors the ‘what’ and ‘how’, but also crucially the ‘why’ and ‘what if?’

By modelling an environment, expertise or task using causal logic, our artificial intelligence safely emulates human decisions at machine-speed, and at an unparalleled scale. Unlike ‘black box’ AI models, our causal AI ensures transparency of each action taken, with every decision fully explained and available for scrutiny and reporting.

You are on the ‘Policy’ node with IP address 192.168.222.1
OK, what should I do next?
You could consider doing a ‘Portscan’ of the device with IP address 192.168.222.1
Why?
Portscan looks for open ports on a device. Open ports can be exploited to get access to a device.

The ‘Wisdom Model’. Because AI is only as good as its training.

// Capturing, sharing and supporting human knowledge and expertise is our biggest challenge with AI. Applying computer science, data science and behavioural science, we have used decades of experience to develop proprietary methods to capture the wisdom of experts and practitioners.

Selecting the very best experts in any organisation or domain, our scientists can rapidly draw out and capture the years of experience and decision logic employees routinely use to successfully perform their job.

For each task we encode a ‘Wisdom Model’ which is a unique, bespoke, knowledge repository. It trains AI agents to execute tasks automatically whilst applying the very best practice, reasoning, and skills. Our agents are constantly learning: as they operate, they improve. Our AI agent’s behaviour can be modified by the user and does not require any coding experience. Furthermore, our 'Wisdom Model' can also be used to ground and validate the output from 'Large Language Models', so improving their trustworthiness.

Human-machine teamwork. Empower and support your people.

// Get more from the knowledge and expertise of your most experienced employees. We believe the combination of human and machine intelligence will be transformational: not an existential threat to humanity or people’s jobs, but a way to make demanding, high-stakes, high-pressure roles easier, safer, more rewarding and more productive.

The lived experience and judgment of high-performing employees is invaluable. Amassed over decades, their knowledge represents an organisation’s most valuable asset.

Our expertise and approach offer the way to re-imagine the workplace. We are dedicated to building human-machine teams by bringing the innate wisdom of people alongside trusted, powerful machine intelligence.

Proof of concept: Machine speed cybersecurity.

// Our first AI agent has successfully been deployed to identify and test vulnerabilities on simulated defence computer networks.

Using our knowledge capture methodology, we built a ‘Wisdom Model’ mirroring the reasoning and actions of expert network penetration testers. Trained in the tactics, techniques and procedures of world-class experts, our AI agent autonomously completes network penetration test exercises, and our lab tests showed that novice pen testers' performance was increased by >300% in terms of their speed and accuracy with the agent operating as their co-pilot.