Let's be honest, generative AI has already messed with quite a few apple carts, but there's still another one waiting in the wings. The commonplace phrase “This is going to end in tears” is one of the most understatements he can make about what might happen in the future.
The waiting apple cart of problems is the management of the business, and especially the execution of the underlying business processes that drive it. These processes are the first and only points of direct interaction with new generation AI implementations. Very often the effects are instantaneous.
Preparing this environment to be ready for further work and to be able to process results in real time is probably one of the next things. of This is the most important step any business can take this year. As Matt Calkins, his CEO and co-founder of process management company Appian, said at the company's Appian World conference in Washington last week, the first thing you need to understand about generational AI is teeth, do not have In quarantine:
We need to spread this basic knowledge to the world. AI is not a standalone technology. It thrives on being accompanied by several other things, such as data and processes. We all know that AI is nothing without data. The problem is, it's not external, general-purpose data that makes AI truly powerful. It's your data. Unfortunately, this is the most difficult to obtain, the least desirable to share or use for training, and probably the most difficult to source and mine internally. To really add value, you need to not only learn this technology, but also find ways to increase productivity within your environment.
With 25 years of experience in developing tools for building and managing digital business processes, Appian already has a history of working on early implementations of AI technology, and with the new developments introduced at the conference, we are now Few companies are able to position themselves as providing business users with secure, advanced process management and the ability to “socialize” and influence the AI systems they produce.
That's why the company is offering a competitive 1-month fixed price to existing users who are starting on the path to integrating AI services into their business process management environments to experience what's possible. We now offer contracts. For those who are not existing customers, the company is touting a new starter kit consisting of what it believes is an ideal basket of products that Calkins hopes will give them a chance for initial success. It is offered at a “very affordable price”. that”.
Protect your data, not just use it
Of course, data is the lifeblood of both artificial AI and business processes, and it is the same data that is inevitably closely linked to both, so it is equally important to utilize the right data and keep it secure. . This is the first and perhaps most important area of danger when using Gen AI services.
Up to a certain point, you can train on general purpose, public data sources, but eventually you will be leveraging company-specific internal data, especially since that data is updated as needed to provide the greatest benefit to your business. bring. And the most recent data is often the most valuable in terms of business results. It also shouldn't become part of the general-purpose data that is most valuable to a company's competitors and available to leak from generative AI services, Calkins said.
AI cannot be allowed to function alone. This is what I call “mixed autonomy”, basically he has an AI driving the car, but he knows that the AI makes a lot of mistakes, so he keeps his hands on the steering wheel. You'll want to keep it. AI does not have human judgment. You can suggest ideas, but we need to make sure that we are in different moments of autonomy. Processes give structure to teams, so AI needs to be part of them. Processes provide that structure and are a series of tasks and handoffs toward an important goal.
A key part of providing this management functionality is Appian's Data Fabric, introduced on the backend in 2022. This provides a virtual database layer that connects all data within an enterprise to a common semantic layer. It can be addressed as if it were local data. You can also optimize the process so that when you repeatedly run a query, the process automatically adjusts to run the query faster, as if relocating the data without running the query. Calkins suggested making the technology respectful of the way enterprise infrastructure is actually orchestrated.
This means that users do not need to train the AI through machine learning. Instead, the data fabric trains the AI algorithms. Whenever a question is asked for an AI service, the data fabric immediately finds the data corresponding to that question, no matter where it resides in the enterprise, and presents it to AI along with the question. However, this means having a decentralized data strategy, which is not suitable for every business. But this means that you can ask questions of your AI service, combined with relevant and up-to-date data, while greatly reducing the risk of internal data leaking into the service's general data pool. Even if it were, Calkins said, it's likely just a bunch of numbers without any relevant context to give it value.
Only the data that corresponds to the authority level of the person asking the question can be provided as an API. So if you care about security, it's a big deal. AI can say weird things, so it's auditable. It's a black box that gives you answers.
This approach also avoids the “elephants never forget” problem, which is another potential hurdle for AI systems. AI never forgets things because it doesn't make its own data stores obsolete. Also, unless a company chooses the very expensive route of having its own isolated AI instance on-premises or using bare metal with a long-term rental at a cloud service provider, a service that allows data stores to be selectively retired. It is not a starter. This gives companies the freedom to decide what information needs to be retired and when.
It also opens up the possibility of working in an AI-agnostic environment without the risk of long-term lock-in. No detailed training is required, and the AI algorithm is asked to process a set of questions against a specific dataset, allowing you to modify the algorithm as needed. This means users can switch to another AI service provider. As Calkins observed, this could improve the bargaining position of companies with AI vendors.
going down the mine
Since Appian acquired Lana Labs in 2021, the potential of process mining has increased, but has not actually been realized. But now that this technology is fully integrated into Appian's suite of tools, we can do something very important: real-time process mining, now available under the name Process HQ. , you should see considerable progress.
So instead of just going back to the process, you can slowly walk through the process to identify and fix process errors, and this can now be achieved in real time while the process is running, allowing users to access any connected data. The source is now accessible. Data fabric drills down to incidents. Users can then measure incidents such as delays or, on the flip side of the coin, see the actual performance of changes or additions to the process. Calkins explained:
Where is the inefficiency? Where is the correlation between time and a particular activity? Whatever variable correlates to it needs to tell you where it's slowing down. Then compare it to the volume of that instance for further analysis. It also recommends certain things you should do to improve efficiency.
Regarding working with AI, he said that a byproduct of using Process HQ could be to help identify and manage hallucinations in AI.
Hallucinations occur when one of two pools of data (the pool of current events that form the questions and the pool of past events that form the comparisons) becomes dangerously thin. Therefore, depletion of any of these can cause hallucinations. If process mining is accelerated in real-time, it could help detect, notice, and fix problems faster.
Process HQ then connects to Elastic Process Execution functionality. This feature is designed to address scalability risks, the associated volatility, and the unpredictability when risks occur. Some businesses know that they are likely to experience a workload spike, but they don't know when or where it will happen. This feature provides a scalability paradigm that manages processing on demand.
my view
There is no doubt that Gen AI is one of the most important and exciting developments to have occurred in the IT industry since the invention of microprocessors, cloud services, and serverless environments. It's no wonder this is probably one of the most dangerous pieces of software ever created. It seems like the conversation is not about what you can do, but what you can't do.
The ability to control and corral monsters will therefore become increasingly necessary in the everyday realities of any business. This is where Appian is firmly pitching its tent. You can imagine that what the company offers doesn't meet the needs of all AI users. For example, I think it could undermine an attempt to run a monster research program needed to discover new drugs, but for the average company, they're looking for valuable answers to typical problems. For business problems, I think Appian has found at least one workable and manageable solution.
