Yes, then again L&D is a huge pain in the neck too. Just as a cop issuing you with a speeding ticket is a pain in the wallet. But then again, you shouldn’t have been speeding! The trick with Soffos is to comply with the mandatory requests as soon as they are discussed, then they won’t be asked of the user again. Those employees who can’t be bothered to follow the rules will be pestered mercilessly, without apology. If it’s the law to know about workplace regulations, that’s an inconvenient but necessary truth.

Yes, if an answer is uncertain, or an employee feels that an answer is wrong, an admin can provide answers and/or uploads documents which contain those answers. This only needs to be done once per query, as AI doesn’t get things wrong (or uncertain) twice.

Obviously, any Software product will have pros and cons, and employees will feedback to managers, who in turn may feedback to Soffos engineers with ongoing improvement suggestions.  But on a day-to-day basis, the answers that Soffos gives will be succinct, approved and correct.

As Soffos is still in Beta, selected users only are being provided with Beta level software, and are not paying anything in return for their assistance and feedback. However, when the full version of Soffos is launched, it will become a SaaS product on an affordable $ per month / per user / seat license basis.

Just as it is every manager’s responsibility to ensure that all policy documentation and regulatory policies and procedures are followed, so it is up to those same people to update the Soffos knowledge system (it’s not really a database in that sense, Soffos ‘understands’ parsed facts as stand-alone knowledge nuggets). 

So, if the governing trade association of USA welders sends new regulatory information to the head of Engineering at Acme, it is that manager’s responsibility to upload that documentation, delete the previous version and mark the replacement as mandatory. The Soffos knowledge representation system will then ‘overwrite’ that knowledge. It’s no different to a manager sending a team-wide email with an attachment entitled ‘new welding specs for this year’.  The good bit via using Soffos is that every employee doesn’t need to read through the whole attachment, they just ask Soffos about the parts they need to know.  However, date stamps on all file uploads will alert managers to when content was last edited or replaced.

There are several implications to this interesting question. Firstly, two-factor log-in and biometric level security protects access at any level from any device.

Even if a hacker could manage to by-pass that, much of the knowledge accessible could be found via the internet or from a trade association policy anyway. The policy on Acme’s pool-car personal mileage or health and safety procedure isn’t sensitive, nor are the specifications of its products. In terms of this non-sensitive knowledge, anyone could find these things out eventually anyway, but Soffos provides it to authorized users in the blink of an eye.

If a laptop was stolen and its password cracked, there’s nothing ‘stored’ that’s accessible via Soffos; so it’s more likely that other original source documents with no relationship to Soffos would be of much more interest to a thief.

In fact, no-one who isn’t an administrator can see anyone’s documents, as documents are removed from the system, only spoken and textual ‘knowledge nuggets’ can answer employees’ questions. But the permissions levels and subject corpus access is set by senior management. Some silos will be accessible by everyone, and that decision is made by relevant administrators.

Obviously, certain silos, containing certain knowledge, must be confidential and only accessible to senior managers. Just as 100 employees of a corporation wouldn’t be shown each-others’ salary details or corporate account statements in a training classroom, neither does Soffos make that sort of information generally available. Instead, Soffos provides instant accurate answers to policy related information: procedures, products, specifications etc. Soffos tells users the ‘need-to-know stuff’ instantly, 24/7, to get the job done efficiently.

It’s also potentially serious when humans get answers wrong. The difference is that AI only ever gets it wrong once, then it won’t ever make the same mistake once it’s re-trained. Also, there are clear safeguards: when Soffos compiles an answer to a question, we use an algorithm to determine how confident Soffos is about the answer. If it’s not sure, it will say so. For example, Soffos could respond “I’m not sure, but it could be…” or “Sarah, I think the answer to your question is x, but it’s not very clear from the knowledge I have access to.”

Human-written and provided material can be ambiguous or certain, depending on the nature of the query. When Sarah from sales asks Soffos the dimensions of the X1 widget, Soffos can only ever provide the knowledge given by the department manufacturing that widget. And if they’ve got it wrong, well that’s a different question. But in short, Soffos learns constantly and certainly, so once corrected, will never give a wrong answer again, unlike humans, who are naturally flawed in that regard.

The intricate detail is extremely complex and a patentable secret, but it’s all about the use of algorithms that combine computational linguistics, contextual memory, deep learning and, crucially, the use of Knowledge Graphs using Extraction Transformation and Load (ETL) processes.

The AI parses words as input from voice or text from resources, so that it ‘understands’ the relationship between words (as concepts or things) not just by simple keyword association (e.g. car = transport) but by well-defined meta-labels, which refer to relationships between language, concepts, objects and questions from an infinite number of possible (and impossible) relationships. A limited set of relation types are used, using global identifiers with unambiguous denotations.  This is combined with semantic ‘Extraction Transformation and Load’ (ETL) processes from structured databases, forming strong associations and disassociations at almost neural speed during the AI’s training.

An example of a Knowledge Graph to draw upon vast amounts of varied information might be a recommendation system for TV shows, movies, songs and albums from an online entertainment provider, to help find relationships between actors, artistes, titles and series. Later, these relationships can be used to predict what customers might like to watch next, equally such a knowledge graph can be used in a corporate knowledge context to answer direct questions.

Let’s say a person watched Star Trek, and wanted recommendations for similar entertainment. A simple keyword database might then suggest ‘Star Wars’, but it wouldn’t ‘understand’ the relationship between the two as sci-fi series, just that they had a common word within their respective titles.

However, if a customer enjoyed the vintage NYPD cop drama series  ‘Hill Street Blues’ and then watched a documentary about Eric Clapton (Life in 12 bars), the knowledge graph could draw on its contextual database to understand both the difference and the association between the two uses of the word ‘blues’ used with a very similar meaning but within a different context (‘12 bar blues’ vs ‘problematic or difficult situations’).

Thus the AI would use its KG to contextualize between the genres of BOTH rock music documentaries AND vintage TV police dramas, whereas a simple keyword association would probably fail hopelessly to provide a meaningful suggestion.

Likewise, in the context of Soffos, the AI would ‘know’ the difference between required criteria for an investment company’s ‘exit strategy requirements’ and the procedure for staff in the event of a fire alarm.

Intentionally, no. The whole point of Soffos is the elimination of time spent searching through documentation for knowledge. This also obviates the potential for loss and theft of commercially sensitive company information when employees are working remotely. 

If Soffos can’t answer a complex question, the query will be added to an admin’s queue, until that person provides answers and/or uploads documents which contain those answers. Soffos keeps asking for knowledge, closing these gaps.  Soffos can also send videos, audio clips and static images as background knowledge.

In theory, if it’s work-related, then yes. Questions can only be answered if managers have uploaded documentation to be parsed and ‘understood’ by Soffos, so that Soffos can answer with a knowledge nugget, rather than a whole chapter and verse response. For example, if Sarah in sales asks about the dimensions of the X1 widget, Soffos will provide a definitive, instant answer. But if Sarah asks whether Jim in accounts is single, and is that Harley Davidson he rides owned outright, she’ll just have to use her charms…