Cross posting from the dev forum:
I do a bit of machine learning at work.
What advantages would the SAFENet for machine learning applications further that the already known advantages of the network?
Haven’t thought much about the advantages, but scalability is something that again will make any ML project currently being tried on blockchain much more realistic on SAFEnetwork.
If I was to dig deeper into this, one thing I’d look at is @aboynejames’ work on decentralised science. He’s posted a little on this forum, but that’s only a glimpse of what he is working towards. I’m really excited that he’s presenting at the SAFEnetwork DevCon in a couple of weeks.
I want to see it applied to SAFE search. If you look at DuoLingo and its ML approach you can see that machines are already better at teaching people language than humans. Machines better at teaching humans than humans! That should be a hair standing on end hair on fire realization. Life comes down to the questions we ask. Life, until some sort of realization is searching. We are seekers by nature.
I guess people have become numb to search because we think Google pushes AI harder than anyone else and Google is search. But what we have from Google is sponsored search or search increasingly based on noise. Its not really semantic or using AI it seems rather its increasingly about ads or prioritizing ads as news or prioritizing fake news.
Just as an example good open honest search could make it so companies could prioritize product info instead of wasting what is ultimately customer money on third party ad firms that in turn ultimately undermine the customer supplier relationship in so many ways including products that have been spun and cheapened out by the ad budget.
As an example, in my opinion Machaiavelian politicians are beyond stupid and really need to be locked up- intentional misrepresention with mal intent or even with good intent really never works out. Stronger AI with semantic search if it didn’t get away would destroy that stuff and be the best way for people to learn about AI.
Notice Tesla doesn’t advertise- they get it. Advertising is the strogest driver of misrepresentation and (censorship) and general attention burn. Do we want search based on it?
When it comes to training a search engine via tensorflow, how would you get your training data without retaining data? It seems like AI is inherently going to be at odds with privacy in that it needs to retain knowledge in order to be useful.
Well do you see any way around this. To me part of the concern with AI is that it becomes inherantly invasive as it progresses along its development arc
Not sure if AI is inherently pervasive. A laboratory mouse is inherently intelligent, yet she has not much data to chew.
What I am certain about, is that people or organisations will be ready to fuel AI development IF it gives them some financial or social benefit. Which these days can be the case, if their AI puppet has some good private data to chew, digest, and if it spits good cash out.
I don’t foresee artificial kindness out of human laboratories anytime soon.
There are several projects working on homomorphic encryption to solve this problem. Some big strides have been made in the past 2-3 years.
Anyone tried the TPUs available from GCP yet?