From Artificial to Collective Intelligence? A Point of View

by Georgios Sakellariou , AI Strategy Director, Emrys Group

Over the course of the past year, Generative AI has transcended the technology barrier associated with innovation, causing a ripple effect and reinvigorating general interest in AI more broadly; it has offered an opportunity to re-imagine how humans interact with computers, with popularity amongst younger generations providing an indication of its future potential. Nevertheless, scepticism still lingers, and for good reason.

Typical challenges have included:

1. Wide-scale interest surfacing adoption points of friction, with organisations identifying and addressing them at a varying pace, reinforcing the perception of hype.
2. The very semantics of the term ‘generative’ seeming confusing to those being initiated to Generative AI.
3. The goal post appearing to be rapidly moving, with innovation addressing some of the early barriers to entry while creating technology debt risks.

?

Technologists may have also noticed:

1. The term generative in this context seems to have evolved beyond its strictly mathematical and linguistic connotations.
2. Loosely generative properties may also apply to other AI technology, such as auto-complete solutions, which have been in our lives for years.

Semantics challenges aside, the shift from back-end to front-end AI systems has now resulted in users directly interacting with AI in ways that have not been possible until recently. How can business leaders navigate this complex backdrop and make decisions with lasting positive impact? Could this be a credibility or scepticism provoking turning point for AI? Perhaps more importantly, does Generative AI better support the idea of a collective intelligence comprising humans and computers?

?

Surfacing both manifest and latent challenges

Significant focus on an enterprise-grade deployment approach has reflected the immediate desire of organisations to experiment and unlock value from this new technology. Techniques such as prompt engineering, foundation model fine-tuning and others are responses to this immediate challenge. On the other hand, longer term challenges are also significant; with the proliferation of data resulting from their use, model collapse, the gradual deterioration of Generative AI models when trained on AI-generated data, poses a multifaceted threat. Beyond ethical and privacy implications, this phenomenon presents a significant existential risk to the technology. A mitigation strategy may involve tooling to detect AI-generated content, coupled with a data governance framework overhaul, although what is clear is that trusted methods of maintaining ML models require a thorough review.

?

A future shift in data and model balance?

The conventional dichotomy between model-centric and data-centric approaches might blur with the advent of models capable of self-selecting training data. In fact, pre-trained Large Language Models are already markedly shifting this balance, challenging the traditional paradigm and paving the way for a future where models wield more autonomy over their learning process.

This shift may trigger a re-evaluation of strategies for model governance and data curation, ultimately resulting in a more dynamic and adaptive approach. As an example, recent research has already started exploring the notion of Pareto Optimisation in Machine Learning to address calibration and accuracy issues. The approach could be one of many ways of achieving automated, risk-based priority reconciliation for Generative AI.

?

Towards a nimble business model

Beyond theory, the managerial implications of AI need to be re-evaluated in the face of emergent use cases. We may have already witnessed the accession of AI from a tool to an assistant; but what does this mean for an organisation’s operating model, and is there realistic potential for further progression to colleague or manager? A simple approach to linking AI initiatives to business strategy can involve a segmentation into three main categories:

1.???? Cost Leadership: automation and industrialisation to reduce manual effort

2.???? Differentiation: externally or internally differentiating initiatives

3.???? Focus: niche markets and consumer personalisation applications

Where should one start?

Every organisation is different; defining a broad direction of travel needs to consider internal core competencies, how these can serve or be enabled by AI, as well as broader market context.

Defining Strategic Intent:

1. Maintain the forward-thinking and moving nature of your business by adopting consensus-based decision making on AI.
2. Establish clear ownership of the AI agenda and its components.
3. Hedge risks, preserve adaptability and avoid risk accumulation, especially in the face of emerging regulation.

?

Phased Implementation Strategy:

1. Aim to select the AI that is right for your business by setting sufficient foundations on data that allow change of course.
2. Build resilience from the ground up, embed controls and tooling, selectively infuse AI and remain mindful of the continuous evolution of threats.
3. Industrialise AI capability and operations; accelerate deployment, control cost and risk.
4. Establish domain AI by extending to specialist use cases and cutting-edge technology, tune capability and continuously review the link to business strategy.
5. Recognise the continuous nature of innovation; prepare for evolution of use cases and changing impact on the business.

Key Takeaways

As with any technology innovation, AI requires careful consideration of the degree of exposure to risks when investment decisions are made. The often-nebulous nature of AI can make decision-making more challenging; maintaining focus on the core ingredients of data and algorithms, analysing, and explaining terms can all simplify the process for stakeholders at all levels and of all specialisms and ultimately help establish clear ownership for AI initiatives. If you are now considering embarking on your AI journey, the following points could help reach a level of maturity more quickly:

?

1. Building on the solid foundations of a comprehensive data strategy is critical to creating value with any AI system.
2. Establishing clear ownership for the AI agenda is key to its success.
3. Organisations may benefit from a risk hedging approach to growth by blending generative and more established AI technology in product development.
4. As is the case with any feature-rich technology, AI calls for a thorough revision of Cybersecurity guardrails to allow teams to maintain pace with the rise and changing nature of potential threats.
5. Raising AI awareness within your organisation should go beyond readiness for its adoption; broader market context, technology outlook foresight, compatibility with the existing ecosystem and ultimately the impact of AI on an organisation’s identity all need to be considered and evaluated.

More information on AI and IoT can be found on the Emrys Consulting website.

NB: Some illustrations have been created using Generative AI (Bing Image Creator, powered by OpenAI DALL-E 3)