**4. Conclusions**

In summary, we can see that the framework orchestrates the intents and associated implementations from different people while keeping the intent and implementation separate.


## **4.1 Challenges and future directions**

At the beginning of the chapter, we established the significance of human AI collaboration, and proceeded to share about our proposed work and its intended contributions towards this goal. In ending this chapter, it is apt to candidly discuss about its potential challenges and associated future directions. To do this, let us expand our view of human AI collaboration beyond the technological lens. Again, what then is human AI collaboration? It is a relationship, fundamentally. Like any successful relationship, trust and communication are crucial. Let us discuss each of these factors:

• Trust. This represents our confidence level in relying on the output from our AI counterpart. How do we ensure that the AI is performing as it should? Is there transparency in the way the AI operates? Are we able to interpret and understand why the AI does what it does? For example, standard-setting

organisations define criteria for many technologies to ensure that compliance guarantees quality, digital security protocol such as SSL ensures the security for internet communications, well-documented manuals aid us in product troubleshooting and maintenance, etc. Every new technology introduced would eventually face questions such as these. Going forward, an area of potential interest lies in the framework integration with proof assistant capabilities. Work plans built upon our proposed framework are type theoretic in nature. The broad idea is therefore to treat every work plan as a theorem to be proven. By proving the theorem (work plan), the strong implication is that the work plan is verified to be working as intended. The ability to frame real world work plans as a mathematical model has desirable benefits in terms of trust, and this is an area which we hope to investigate more deeply.

• Communication. This represents how human and AI convey and exchange information. While our proposed work contributes towards a facet of human AI collaboration to enable the description and orchestration of intents across a network of humans and machines, it is targeted and focused as is the nature of research. As an analogy, while networking protocols enable the exchange of information over the Internet, it does not inherently make information easily searchable by users. For this, technologies such as search engine come into play. Switching back to our context here, an interesting aspect of communication (beyond our proposed work) would be to consider how these intents (along with its associated implementations and data) can be made discoverable and reusable by others.

Naturally, our discussion here is by no means exhaustive. It is our intent and hope that our proposed work and discussion contributes towards and catalyse future discussions in the research community for the continued advancements of human AI collaboration and ultimately, towards the future of a collaborative human AI society.
