Hey everyone, been a LONG while since I posted on here!  Hope you are all well.

Been very very busy these past few months and thought it was about time I dropped by and made a post as what I’ve been working on is kinda relevant to this forum.

Back in Aug/Sept, some of you may remember I was working on an AI/Chat called ALF.  The intention was to implement true NLP and “reasoning” into that AI for a number of projects I wanted to do, not just a Chatbot.

Well, the first commercial use of that AI has gone live this weekend, you can’t chat with it yet, but it powers a bunch of services behind my startup.

The endevour was to create a product search engine, with the goal to become the largest and most comprehensive in the UK, and later other territories.  One of the requirements of this project was that it had to incorporate a lot of social elements into it, for example where you can interact closely with other people that have similar interests in particular products the same as you.

One feature in particular is very powerful, and is powered by the ALF AI.  Once you have signed up, ALF “follows” you around the internet (stalker AI anyone? ) and learns about you…what things you like, what movies, what music etc etc and will create a suggestions list for you of items that you might like. 

Additionally, if you sign up with Twitter or FB accounts, he utilizes them for your suggestions, reads your statuses and other available information.  We also collate public information on your friends, this allows us to build suggestions for your friends too. Example, if a close friend has a birthday coming up, he will be able to provide you with targetted, accurate suggestions of what that friend might like.

Thats just a couple of examples, but the idea runs a lot further.

ALF also powers the search engine to a degree, we perform regular search via keyword lookups and other regular signals.  The results are then passed to ALF for tuning to you depending on what he has learned about you.

So if you liked racing games and had an XBox, and a friend liked adventure games and had a Wii, if you were both to type in “Video games”, both would get the relevant search results.

I dont want to drag this post on too long, but the above is the tip of the iceberg really.  The next development is true NLP and “understanding” in the search field, so you could for example search for “Motherboards compatible with Opteron processors” etc

If you would like to try it out, the address is http://www.koliko.co.uk and you can use the beta code of 513bd286.  Obviously we are in beta and killing gremlins daily

Any questions, thoughts please post them, Im eager to hear people opinions.

Thanks

Dan

Nice to hear about ALF.
Hope to chat with it soon, I am very curious abou how will it menage a conversation.
But I have some doubts about the current site>
It takes my preferences about my searches or about the things I buy?
Will it be available as an of-line personal assintent in a second moment?
I remember that some of the targets to ALF were :Being able to have conversations, recalling past conversations and make associations with context.
You also said about an AI context of time and image recognizing, how will it work in an e-commerce site?
You also said something about ALF recognizing permanet or impermanet states. Will ALF learn it form users or all informations must be programmed ?
But what is the difference of today’s diference between ALF and Google? Coz both will show my preferences after some researches in www.
Maybe testing it on facebook, but does it already work in this beta version?

Well at present in the current service there is no conversation with it.

The way he works is different to Google, in the way that he reads about you…for example if you posted on Facebook

“Johnny Depp is great!”

ALF would read that, and deduce that you like Johnny Depp, he is an actor, so you probably like his films, you probably like movies in general, and would give films of the genre that he has stared in some priority.

From that ALF can then hone suggestions to you.

Google looks at where you have been, ALF looks at what you have said, what you have shared, what you have liked.  Theres a big different between the 2.

In the future, ALF will have many more features on the service, one being an AskALF assistant where you can ask questions on products in a true form.  ALF will attempt to find the correct answer and converse with you, maybe asking you questions to further determine the context and nature of what it is you are looking for.

Oh! Sorry I was actualizing the post and you already answered.
It seems great to me. So I presume Facebook interfece is completely funcitional?
I’ll try it.

yeah its in beta, we launched today, so there still may be a few gremlins here and there, but thats the idea of a beta, to help you find them

If you want to share the link with anyone, you can use the code I gave in the original post so they can sign up and try it out.

Also, we have launched only in UK, anyone can sign up, but the products it finds for you etc will be from UK merchants.

There are plans to expand of course, but baby steps is key

Thank you !

One question about Twitter account.
Is it ALF on Twitter on a human user? I saw it twited twice. Does ALF in twitter also saves the preferences?
I took a look at my friends preferences and despite one or another incongruence, it works fine.
What I mean about incongruences: I and some of my friends are Michael Jackson’s fans, so ALF sent recomendations about his cds, but also about Michael Bolton and Michael Bouble. I am sure I never pressed the like FB button to Michael Bouble or Michael Bolton. So, in this case, ALF didn’t “understood” my likes and dislikes, but just gave me some other “Michaels”. I also asked about it to my friends.
In another case, one of my friends never talked about Harry Potter ( because of religious motivations) and HP was it’s indications for her… So some improvements have to be done.

I saw some AI on Twitter ( from http://www.a-i.com/). @twthal is one of them. It seems it’s used basicaly to make divulgation of scientific new on Twitter, but sometimes it is on for chat.

You’re using a very intresting aproach and I am happy if I can help.

Hi Fatima,

Thanks for trying it out and for your support.

Of course, he is still quite early in development, so there are of course many improvements still to be made.  He makes assumptions that sometimes are not correct, the Micheal Jackson - Bouble is one of them. 

As more people sign up and use it, there will be more data available for him to see trends, at the moment he thinks there is a trend between those 2, when really its just that they are musicians, with more data, the trends will be more focused and results will get better and better.

Twitter atm is human, although I do want ALF to have social interactions too in the future.

I checked out your web site and can’t figure out where the actual AI is being used? So from what I understand, ALF is a shopping agent of sorts that follows and tracks your shopping preferences? As it records your shopping habits it assists in making product purchasing decisions?

My questions is, other than tracking your social and shopping habits, where is the AI actually being utilized? I would so I would assume at some point you will be able to converse with ALF and it will help you shop with a comparison of products?

I am interested in your project because my own bot in development has a similar functionality in that Marie will gather product recommendations during the course of a conversation.

Is that what you envision ALF’s capabilities to be?

Hi Laura,

Its used in the search, and for product suggestions for you and friends when you are a member.

You are fairly accurate with your understanding, ALF acts as an assistant, recommending things you may like from information he’s learned about you, both on Koliko, from your activities, and also external accounts that you have linked.

For example, he reads you Facebook account if you have it connected, not just your like, but also any statuses that you posted.

So if you posted “My cooker is broken, I need a new one” or similar, shortly after, cookers would appear in your recommendations.

The current version is early on, and is just a simple implementation of that behaviour that I want to expand on.

Aside from that, I would like him to become the store assistant, with an “Ask Alf” button that pops up a chat window where you can ask him about products, ask questions and have a general conversation should you wish.

Past that, I also want him to be active in social media sites.

So again, if you are linked to Koliko with your FB account for example, I’d like ALF to be able to engage with users on FB directly, comment on wall posts etc.  Not as a sales driver, just because it is cool!  The hope being that people will associate ALF with Koliko and shopping, then come to the site when they want to shop

Hi Dan,

Thanks for your clarification

So your bot/agent crawls the user’s social network and data mines the user’s posts as well as their friends for relevant keywords/phrases that is then parsed and used to assist in the user’s shopping experience, making gift suggestions and product recommendations?

Sounds interesting and I look forward to your website as it matures. Your market just for the UK, correct?

Hi Laura,

You got it

It doesnt look for keywords tho, it actually reads the sentence, parses it, breaks it down via NLP, then attempts to construct the concept of the sentence.

Just UK at the moment, but plans to expand to Europe and the US once we are certains that its standing ok in the UK.

D

Dan, what you’re describing sounds like the culmination of the worst fears of both online privacy advocates and anyone who’s ever been annoyed by a pop up ad. I foresee the only way to convince people to sign up for a service like this is enticing them with the chatbot aspect of your project. And heaven forbid the chatbot publicly allude to some potentially embarrassing information gleaned from someone’s purchasing habits.

CR,

I had thought about the privacy issues involved as well. Having to sign up for and give permission to access personal social networking accounts is the kill buzz here. I agree that in the course of chatting with a bot in a co-browsing environment, if relevent products or services would display that would be a much more acceptable way and less intrusive to the user’s privacy than a background stealth approach.