Webinar | AI and IP – your questions answered
16th February 2018 |
The Cipher Webinar, AI and IP – a meeting of minds gained a record number of attendees and generated dozens of the questions. We are sharing the full list below, answered by Nigel and Steve and Julian Nolan, CEO of Iprova.
Please share your thoughts in the comments section below. Let’s keep the conversation going!
1.What does Aistemos do? Is it the same company as Cipher?
Aistemos is our company name (aistemos.com) and Cipher is the product name for our range of patent analytics software and solutions (cipher.ai).
2.In order to develop the patent asset class, patents need to be granted within a reasonable time frame – two or three years after filing. Do you agree? How do we get Patent Offices to work in this timeframe?
We think speed of grant is important, but not a precondition to the recognition of patents as an asset class. We do think that transparent and accurate information about ownership is fundamental.
3.Any traction with government cooperation on ORoPO?
ORoPO has the support of WIPO, KPO, UKIPO, EPO and USPTO. Everyone wants to solve the ownership issue, but there are currently insufficient incentives to make patent owners participate.
4.Do you have an example of a patent application “written” by an AI or an invention made by an AI?
No, but there are many companies currently offering machine-assisted patent drafting.
5.Does AI and Blockchain transparency effectively make the patent system obsolete?
No. Greater transparency (IPwe’s proposed Blockchain) and improved understanding (Cipher’s AI approach to patent analytics) will make the patent system function as intended.
6.How does Cipher evaluate patents and identify development trends, opportunities and impediments?
Cipher takes a holistic view of patent information. It aggregates patent, licensing, litigation, SEP and cost data and then uses AI and machine learning to cluster and compare companies and technologies. By leaving behind artificial constraints such as keywords and Patent Office codes, Cipher can focus on delivering analysis that fosters collaboration between all the teams who benefit from understanding the innovation landscape. For more information or a demonstration please contact email@example.com.
7. Artificial intelligence can replace Lawyers?
A question or a statement? AI can replace aspects of all jobs that are mechanical and repeatable. That impacts all of us, and lawyers do not have a special exception. Cipher performs about 60% of the work associated with patent due diligence – previously labour intensive and boring.
8. What are the proxies, other than citations (which are not clear enough) which give an indication of value for a standard essential patent (SEP), so that classifiers can be built to propose, with sufficient probability, whether a patent would become SEP?
We are currently working on the right approach SEP classifiers. Classifiers are built from positive (SEPs) and negatives (related, but non-SEPs) examples.With these inputs, our software can look at all metadata e.g. words (text/claims), codes, and citations to determine the likelihood of a patent being a SEP. While this approach doesn’t get to the accuracy of detail expert review, we believe that this type of objective and repeatable process, with measurable precision and recall, may be better than human sampling and speed-reading, which are typically the reality. The recent SEP litigation in the EU and the US establishes that this is widely inconsistent.
9. Transparency and ownership is a clear win-win situation. Talking with computer scientists, a number of people claim that is an easy problem to solve and can’t understand why this hasn’t happened yet in the IP domain. What do you think are the barriers for adopting either Blockchain or AI to overcome this particular problem and what is the key action or milestone that needs to happen in order for these to enable transparency?
Patents will not be a functional asset class until there is a reliable way to establish who own’s what. Refer to the ORoPO Report (2015). Fixing the problem requires cooperation between 100 countries (or at least the 5 IP offices ) or a voluntary solution (which is the ORoPO approach). The only barrier to progress in the right incentive. The US attempted the “stick” (their Attributable Ownership initiative). IPwe is considering “carrots”. Blockchain will have a role to play to ensure that the information is verified and stored – the hard part is getting owners to disclose what they own. Hard to believe.
10. Does Iprova use AI to design and even produce new articles?
JN: No, whilst there have been attempts at this in the past (for example, using genetic algorithms by HP’s GP Lab), Iprova does not use technologies such as AI to create new product designs. We are focused on creating disruptive and yet patentable inventions which can directly drive new value creating products and services for our customers.
11.Automatic Discovery of Concepts is a current trend. If I refer you to Karpathy’s blog on Recurrent Neural Networks, which can generate new text from learning, to what extent do you think this can be applied to patents in both automatic concept generation (in novelty and inventive step) and automatic drafting of patents?
JN: RNNs, when used as a language model as in Karpathy’s blog, are great to model grammatical and semantic constraints in text but cannot easily (and do not really have a strong incentive to) learn the complex interactions required for automatic concept generation. Relying on one unique model to generate novel and inventive concepts would first require the design of new learning tasks, datasets, and algorithms, giving models more incentive to learn the right underlying patterns.