Artificial Intelligence research increases, and we get new algorithms surpassing academic performance benchmarks on an almost weekly basis. The real effect of them on people’s lives, however, lies on what we wold like to call the “AI Supply Chain.”
And there is few who disagree with the fact that the most important part of a value chain is the consumer. Buyers’ decisions ultimately define the shape and fate of a value chain, and AI is no exception.
This explains a key and perennial point of contention. Companies want your money, so they need your trust. They will try to convince you revealing as little information as possible. There is a fine line between what seems reasonable or acceptable, and what amounts to deception and illegal marketing practices.
This could not ring more true in today’s AI marketplace.
Before moving forward, we are defining “AI” very loosely, because the definition of AI itself is loose and encompassing generous. As we previously attempted to define it:
AI Shopping list, item one: Your problems
First of all, make sure that when considering your problem you are not retroactively justifying the acquisition of a technology.
- What are critical problems in your organization that AI could play a role?
- Are there clear inefficiencies that your organization could address through AI?
- Are employees facing routine and predictable tasks that could be automated?
- Do you have assets, data in particular, that are being underutilized?
AI Solutions
- What is the ideal state of AI implemented in your organization? In other words, how will your organization look like when an AI solution implemented?
- How will the AI be integrated into your existing technological solutions bundle? Can it learn from data and processes in existing systems?
- And how will it be integrated into your operations, process, management and employee workflows?
- What are the quantifiable and expected benefits of the AI? Does it increase revenues, reduce costs, makes processes measurably more efficient?
- What are your expectations in terms of time and impact? How long are you willing to wait, for what level of technological transformation? While it’s true that these technologies take some time to “learn” to start delivering results, often vendors use this to justify poor performance for needlessly long times.
- Is the solution “sustainable”? Not referring to environmental concerns (although we probably should), but the ability for the solution to be useful long after the provider engagement is over.
AI Technological Inventory
- Is there an AI already available that would take you from your problem state to your desired solution state?
- Are the solutions available specific to your industry or field? Can a general-purpose AI be useful?
- How do the AI solutions fare compared to non-AI solutions?
- How much technical involvement does the product need from your organization?
- How much workplace training does the AI require to be used to the best of its capabilities?
- Which physical resources does the AI use? How does it use storage and computing power, and who is footing the bill for those?
- What are the visible but hard-to-quantify benefits of AI? Does it positively impact employees’ experiences? Does it take teams to new levels of understanding?
- AI is not out there, is your organization capable of carrying out an AI development project for internal, possibly external uses?
Data
- Do you have data?
- If you have data, doo they need organizing, cleaning and make it readable to humans and machines?
- If you do not have data, do you have records that could become data (i.e. folders) and need digitalization and processing?
- Do you need external data sources?
- Do you need to put in place data generation, gathering or collection processes?
- Will you remain in control of the data and continue to benefit from extracting value out of it?
Risk management
- How can you verify the impact of AI in your current projects and operations?
- Does the AI give you the tools to properly monitor its function, outputs and use of resources?
- How easy is to fix outstanding issues in the system? If needed, how are the provider’s SLAs?
- How does your AI address security issues, design flaws and bugs, permissions, and privacy concerns?
- How will customers and the community react to know you are taking advantage of AI? How will the product address ethical and social concerns?
- Which new risks will the AI bring to the organization? Are they worth having? (And again a nod to the environment.)
Off to the AI megastore
The following survey shows just a small sample of what is available out there. We are still at a level where education-specific AI is yet to materialize. Which is another way to say that the AI education consumer is underdeveloped.
Case use aisles
- Drawing insight from large sets of data.
- Personalize user experiences
- Optimize team and employee workflow
- Optimize resource allocation or maximize use of installed capacity
- Forecast future events
- Facilitate research, for example through semantic search
- Lower risk, uncertainty or employee anxiety
- Alert of potential issues
- Strengthen quality and service, enhance customer experience
- General support, real-time support (mostly assistants and
- chatbots)
- Educate about how AI works: Make uncertainty manageable, algorithm-based decision making understandable
- Deliver employee training
- To generate insights about possible future avenues of growth, innovation, or value differentiation
- Process language for education (grammar, etc.), assessment and plagiarism, translation or other
Here are some examples for the categories listed. We hope to expand on them over the coming months.
- Intelligence. Some universities have massive information repositories dating back years. Obtaining valuable information is challenging, and Google will just won’t cut it. There is perhaps no better example out there than iSeek.ai
- Product development. While others have access to vast (and expensive) third-party repositories, and an investment in an AI product would actually make the initial expense more valuable. Example: InnovationQ Plus
- Automation. A more labor intensive side of the equation entails the role of people in helping train models, tag data, or completing jobs in the hopes that they will eventually be taken over by machines. Example: Appen (formerly Figure Eight)
- Language (Natural Language Processing, NLP). Some general purpose translation assistants are already available, including on Microsoft Skype. A similar technology on eBay is reportedly responsible for improving the efficiency of trade and increasing the volume in ways that would otherwise not be possible (PDF) And of course, we could not pass up mentioning Poodll’s plugin family for language learning and speaking practice, featuring text-to-speech and speech-to-text capabilities.
- Education-Sandbox. Weights & Biases’ value proposition is hard to grasp. But they do shed some light on the specifics of AI Project Management, and how they can be radically different in terms of risk and expectations.
- Pattern matching. Technologies like SIM2REALAI grab high-resolution behavioral patterns of people which are then used to train machines performing basic tasks. The technology, however, can be easily repurposed to create a “data footprint” of a learner’s behavior during a learning activity.
It’s worth noting that along the evaluation and risk management operations, adding some “pwning surveillance” should come in hand. Increasingly, some applications are the target of hackers who steal personal and payment information. Some companies can detect these attacks while others only realize when their users’ data ends up in the dark web.
Not on sale… yet?
Here we list technologies from learning organizations, or tackling learning issues inside organizations. Their availability to the public, and arguably it’s long-term quality, is not guaranteed.
- University of Alabama at Birmingham’s “Intrusive advising assistants” (see video) track student behavior and anticipate issues.
- Cybersecurity AI is a special case, as it is often funded by government or public agencies, and it’s in their best interest to make developments available for anyone. Such is the case of “adversarial AI” research, advanced by DARPA and UC Berkeley, among others.
More Education in AI Pulse
A bold plan to educate 1% of the (admittedly small) population of Finland on practical AI, to eventually expand to a is underway. The initiative, of government origin, has locked in key advocacy, academic and private players, increasing hopes of high adoption across the population. Not Finnish? No problem! Take the “Introduction to AI” course by yourself.
And is AI bringing a new age of algorithmic cheating? TalkToTransformer takes advantage of the open source GPT AIs for text generation.
